""" Layout-Preserving PDF Generation Service Generates PDF files that preserve the original document layout using OCR JSON data """ import json import logging import re from pathlib import Path from typing import Dict, List, Optional, Tuple, Union from datetime import datetime from reportlab.lib.pagesizes import A4, letter from reportlab.lib.units import mm from reportlab.pdfgen import canvas from reportlab.pdfbase import pdfmetrics from reportlab.pdfbase.ttfonts import TTFont from reportlab.platypus import Table, TableStyle, SimpleDocTemplate, Spacer from reportlab.platypus import Image as PlatypusImage from reportlab.lib import colors from reportlab.lib.enums import TA_CENTER, TA_LEFT, TA_RIGHT from reportlab.platypus import Paragraph from reportlab.lib.styles import getSampleStyleSheet, ParagraphStyle from PIL import Image from html.parser import HTMLParser from app.core.config import settings # Import table column corrector for column alignment fix try: from app.services.table_column_corrector import TableColumnCorrector TABLE_COLUMN_CORRECTOR_AVAILABLE = True except ImportError: TABLE_COLUMN_CORRECTOR_AVAILABLE = False TableColumnCorrector = None # Import text region renderer for simple text positioning try: from app.services.text_region_renderer import TextRegionRenderer, load_raw_ocr_regions TEXT_REGION_RENDERER_AVAILABLE = True except ImportError: TEXT_REGION_RENDERER_AVAILABLE = False TextRegionRenderer = None load_raw_ocr_regions = None # Import UnifiedDocument for dual-track support try: from app.models.unified_document import ( UnifiedDocument, DocumentElement, ElementType, BoundingBox, TableData, ProcessingTrack, DocumentMetadata, Dimensions, Page, StyleInfo ) UNIFIED_DOCUMENT_AVAILABLE = True except ImportError: UNIFIED_DOCUMENT_AVAILABLE = False UnifiedDocument = None logger = logging.getLogger(__name__) class HTMLTableParser(HTMLParser): """Parse HTML table to extract structure and data""" def __init__(self): super().__init__() self.tables = [] self.current_table = None self.current_row = None self.current_cell = None self.in_table = False def handle_starttag(self, tag, attrs): attrs_dict = dict(attrs) if tag == 'table': self.in_table = True self.current_table = {'rows': []} elif tag == 'tr' and self.in_table: self.current_row = {'cells': []} elif tag in ('td', 'th') and self.in_table and self.current_row is not None: colspan = int(attrs_dict.get('colspan', 1)) rowspan = int(attrs_dict.get('rowspan', 1)) self.current_cell = { 'text': '', 'is_header': tag == 'th', 'colspan': colspan, 'rowspan': rowspan } def handle_endtag(self, tag): if tag == 'table' and self.in_table: if self.current_table and self.current_table['rows']: self.tables.append(self.current_table) self.current_table = None self.in_table = False elif tag == 'tr' and self.current_row is not None: if self.current_table is not None: self.current_table['rows'].append(self.current_row) self.current_row = None elif tag in ('td', 'th') and self.current_cell is not None: if self.current_row is not None: self.current_row['cells'].append(self.current_cell) self.current_cell = None def handle_data(self, data): if self.current_cell is not None: self.current_cell['text'] += data.strip() + ' ' class PDFGeneratorService: """Service for generating layout-preserving PDFs from OCR JSON data""" # Font mapping from common fonts to PDF standard fonts FONT_MAPPING = { 'Arial': 'Helvetica', 'Arial Black': 'Helvetica-Bold', 'Times New Roman': 'Times-Roman', 'Times': 'Times-Roman', 'Courier New': 'Courier', 'Courier': 'Courier', 'Calibri': 'Helvetica', 'Cambria': 'Times-Roman', 'Georgia': 'Times-Roman', 'Verdana': 'Helvetica', 'Tahoma': 'Helvetica', 'Trebuchet MS': 'Helvetica', 'Comic Sans MS': 'Helvetica', 'Impact': 'Helvetica-Bold', 'Lucida Console': 'Courier', 'Palatino': 'Times-Roman', 'Garamond': 'Times-Roman', 'Bookman': 'Times-Roman', 'Century Gothic': 'Helvetica', 'Franklin Gothic': 'Helvetica', } # Style flags for text formatting STYLE_FLAG_BOLD = 1 STYLE_FLAG_ITALIC = 2 STYLE_FLAG_UNDERLINE = 4 STYLE_FLAG_STRIKETHROUGH = 8 def __init__(self): """Initialize PDF generator with font configuration""" self.font_name = 'NotoSansSC' self.font_path = None self.font_registered = False self.current_processing_track = None # Track type for current document self._register_chinese_font() def _register_chinese_font(self): """Register Chinese font for PDF generation""" try: # Get font path from settings font_path = Path(settings.chinese_font_path) # Try relative path from project root if not font_path.is_absolute(): # Adjust path - settings.chinese_font_path starts with ./backend/ project_root = Path(__file__).resolve().parent.parent.parent.parent font_path = project_root / font_path if not font_path.exists(): logger.error(f"Chinese font not found at {font_path}") return # Register font pdfmetrics.registerFont(TTFont(self.font_name, str(font_path))) self.font_path = font_path self.font_registered = True logger.info(f"Chinese font registered: {self.font_name} from {font_path}") except Exception as e: logger.error(f"Failed to register Chinese font: {e}") self.font_registered = False def _detect_content_orientation( self, page_width: float, page_height: float, ocr_data: Dict ) -> Tuple[bool, float, float]: """ Detect if content orientation differs from page dimensions. This handles cases where a document is scanned in portrait orientation but the actual content is landscape (or vice versa). PP-StructureV3 may return bounding boxes in the "corrected" orientation while the image remains in its scanned orientation. Args: page_width: Declared page width from image dimensions page_height: Declared page height from image dimensions ocr_data: OCR data dictionary containing bounding boxes Returns: Tuple of (needs_rotation, adjusted_width, adjusted_height) - needs_rotation: True if page orientation should be swapped - adjusted_width: Width to use for PDF page - adjusted_height: Height to use for PDF page """ # Find max content bounds from all regions max_x = 0 max_y = 0 all_regions = [] # Collect regions from various sources if 'text_regions' in ocr_data and isinstance(ocr_data['text_regions'], list): all_regions.extend(ocr_data['text_regions']) if 'layout_data' in ocr_data and isinstance(ocr_data['layout_data'], dict): elements = ocr_data['layout_data'].get('elements', []) if elements: all_regions.extend(elements) if 'images_metadata' in ocr_data and isinstance(ocr_data['images_metadata'], list): all_regions.extend(ocr_data['images_metadata']) for region in all_regions: try: bbox = region.get('bbox') if not bbox: continue # Handle different bbox formats if isinstance(bbox, dict): # BoundingBox object format max_x = max(max_x, float(bbox.get('x1', bbox.get('x0', 0) + bbox.get('width', 0)))) max_y = max(max_y, float(bbox.get('y1', bbox.get('y0', 0) + bbox.get('height', 0)))) elif isinstance(bbox, (list, tuple)): if len(bbox) >= 4 and isinstance(bbox[0], (int, float)): # [x1, y1, x2, y2] format max_x = max(max_x, float(bbox[2])) max_y = max(max_y, float(bbox[3])) elif isinstance(bbox[0], (list, tuple)): # Polygon format [[x, y], ...] x_coords = [p[0] for p in bbox if len(p) >= 2] y_coords = [p[1] for p in bbox if len(p) >= 2] if x_coords and y_coords: max_x = max(max_x, max(x_coords)) max_y = max(max_y, max(y_coords)) except Exception as e: logger.debug(f"Error processing bbox for orientation detection: {e}") continue if max_x == 0 or max_y == 0: # No valid bboxes found, use original dimensions return (False, page_width, page_height) logger.info(f"內容邊界偵測: max_x={max_x:.1f}, max_y={max_y:.1f}, " f"page_dims={page_width:.1f}x{page_height:.1f}") # Calculate how much content extends beyond page boundaries x_overflow = max_x / page_width if page_width > 0 else 1 y_overflow = max_y / page_height if page_height > 0 else 1 # Check if content significantly exceeds page dimensions in one direction # This suggests the content is in a different orientation OVERFLOW_THRESHOLD = 1.15 # Content extends >15% beyond declared dimensions if x_overflow > OVERFLOW_THRESHOLD and y_overflow <= 1.05: # Content is wider than page but fits in height # This suggests portrait image with landscape content logger.warning(f"偵測到內容方向可能與頁面不符: " f"x_overflow={x_overflow:.2f}, y_overflow={y_overflow:.2f}") # Check if swapping dimensions would help # If max_x fits better in page_height, swap if max_x <= page_height * 1.05: logger.info(f"建議頁面旋轉: {page_width:.1f}x{page_height:.1f} -> " f"{page_height:.1f}x{page_width:.1f}") return (True, page_height, page_width) else: # Content still doesn't fit, just scale to fit content logger.info(f"內容超出頁面邊界，調整頁面大小以容納內容") return (False, max_x * 1.02, page_height) elif y_overflow > OVERFLOW_THRESHOLD and x_overflow <= 1.05: # Content is taller than page but fits in width # Less common - landscape image with portrait content logger.warning(f"偵測到內容方向可能與頁面不符 (高度溢出): " f"x_overflow={x_overflow:.2f}, y_overflow={y_overflow:.2f}") if max_y <= page_width * 1.05: logger.info(f"建議頁面旋轉: {page_width:.1f}x{page_height:.1f} -> " f"{page_height:.1f}x{page_width:.1f}") return (True, page_height, page_width) else: logger.info(f"內容超出頁面邊界，調整頁面大小以容納內容") return (False, page_width, max_y * 1.02) # No orientation issue detected return (False, page_width, page_height) def _parse_color(self, color_value) -> Tuple[float, float, float]: """ Parse color value to RGB tuple. Args: color_value: Color as hex string (#RRGGBB), RGB tuple, or color name Returns: RGB tuple with values 0-1 for ReportLab """ if not color_value: return (0, 0, 0) # Default to black try: # Handle hex color (#RRGGBB or #RGB) if isinstance(color_value, str) and color_value.startswith('#'): hex_color = color_value.lstrip('#') # Expand short form (#RGB -> #RRGGBB) if len(hex_color) == 3: hex_color = ''.join([c*2 for c in hex_color]) if len(hex_color) == 6: r = int(hex_color[0:2], 16) / 255.0 g = int(hex_color[2:4], 16) / 255.0 b = int(hex_color[4:6], 16) / 255.0 return (r, g, b) # Handle RGB tuple or list elif isinstance(color_value, (tuple, list)) and len(color_value) >= 3: r, g, b = color_value[0:3] # Normalize to 0-1 if values are 0-255 if any(v > 1 for v in [r, g, b]): return (r/255.0, g/255.0, b/255.0) return (r, g, b) except (ValueError, TypeError) as e: logger.warning(f"Failed to parse color {color_value}: {e}") # Default to black return (0, 0, 0) def _map_font(self, font_name: Optional[str]) -> str: """ Map font name to PDF standard font. Args: font_name: Original font name Returns: PDF standard font name """ if not font_name: return 'Helvetica' # Direct lookup if font_name in self.FONT_MAPPING: return self.FONT_MAPPING[font_name] # Case-insensitive lookup font_lower = font_name.lower() for orig_font, pdf_font in self.FONT_MAPPING.items(): if orig_font.lower() == font_lower: return pdf_font # Partial match for common patterns if 'arial' in font_lower: return 'Helvetica' elif 'times' in font_lower: return 'Times-Roman' elif 'courier' in font_lower: return 'Courier' # Default fallback - use NotoSansSC for CJK support if registered if self.font_registered: logger.debug(f"Font '{font_name}' not found in mapping, using {self.font_name} for CJK support") return self.font_name else: logger.debug(f"Font '{font_name}' not found in mapping, using Helvetica") return 'Helvetica' def _apply_text_style(self, c: canvas.Canvas, style_info, default_size: float = 12): """ Apply text styling from StyleInfo to PDF canvas. Args: c: ReportLab canvas object style_info: StyleInfo object or dict with font, size, color, flags default_size: Default font size if not specified """ if not style_info: # Apply default styling c.setFont('Helvetica', default_size) c.setFillColorRGB(0, 0, 0) return try: # Extract style attributes if hasattr(style_info, '__dict__'): # StyleInfo object font_family = getattr(style_info, 'font_name', None) font_size = getattr(style_info, 'font_size', default_size) color = getattr(style_info, 'text_color', None) font_weight = getattr(style_info, 'font_weight', 'normal') font_style = getattr(style_info, 'font_style', 'normal') # Legacy flags support flags = getattr(style_info, 'flags', 0) elif isinstance(style_info, dict): # Dictionary font_family = style_info.get('font_name') font_size = style_info.get('font_size', default_size) color = style_info.get('text_color') font_weight = style_info.get('font_weight', 'normal') font_style = style_info.get('font_style', 'normal') # Legacy flags support flags = style_info.get('flags', 0) else: # Unknown format, use defaults c.setFont('Helvetica', default_size) c.setFillColorRGB(0, 0, 0) return # Map font name base_font = self._map_font(font_family) if font_family else 'Helvetica' # Determine bold and italic from font_weight/font_style (preferred) or flags (legacy) is_bold = font_weight == 'bold' if font_weight else bool(flags & self.STYLE_FLAG_BOLD) is_italic = font_style == 'italic' if font_style else bool(flags & self.STYLE_FLAG_ITALIC) # Apply bold/italic modifiers if is_bold or is_italic: if is_bold and is_italic: # Try bold-italic variant if 'Helvetica' in base_font: base_font = 'Helvetica-BoldOblique' elif 'Times' in base_font: base_font = 'Times-BoldItalic' elif 'Courier' in base_font: base_font = 'Courier-BoldOblique' elif is_bold: # Try bold variant if 'Helvetica' in base_font: base_font = 'Helvetica-Bold' elif 'Times' in base_font: base_font = 'Times-Bold' elif 'Courier' in base_font: base_font = 'Courier-Bold' elif is_italic: # Try italic variant if 'Helvetica' in base_font: base_font = 'Helvetica-Oblique' elif 'Times' in base_font: base_font = 'Times-Italic' elif 'Courier' in base_font: base_font = 'Courier-Oblique' # Apply font and size actual_size = font_size if font_size and font_size > 0 else default_size try: c.setFont(base_font, actual_size) except KeyError: # Font not available, fallback logger.warning(f"Font '{base_font}' not available, using Helvetica") c.setFont('Helvetica', actual_size) # Apply color rgb_color = None if hasattr(style_info, 'get_rgb_color'): # Use StyleInfo method if available rgb_color = style_info.get_rgb_color() elif color is not None: # Parse from extracted color value r, g, b = self._parse_color(color) rgb_color = (r, g, b) if rgb_color: # text_color is in 0-255 range, convert to 0-1 for ReportLab r, g, b = rgb_color if any(v > 1 for v in [r, g, b]): r, g, b = r/255.0, g/255.0, b/255.0 c.setFillColorRGB(r, g, b) else: c.setFillColorRGB(0, 0, 0) # Default black except Exception as e: logger.error(f"Failed to apply text style: {e}") # Fallback to defaults c.setFont('Helvetica', default_size) c.setFillColorRGB(0, 0, 0) def load_ocr_json(self, json_path: Path) -> Optional[Dict]: """ Load and parse OCR JSON result file Args: json_path: Path to JSON file Returns: Parsed JSON data or None if failed """ try: with open(json_path, 'r', encoding='utf-8') as f: data = json.load(f) logger.info(f"Loaded OCR JSON: {json_path.name}") return data except Exception as e: logger.error(f"Failed to load JSON {json_path}: {e}") return None def _get_image_path(self, element) -> Optional[str]: """ Get image path with fallback logic. Checks multiple locations in order: 1. element.content["saved_path"] - Direct track saved path 2. element.content["path"] - Legacy path 3. element.content["image_path"] - Alternative path 4. element.saved_path - Direct attribute 5. element.metadata["path"] - Metadata fallback Args: element: DocumentElement object Returns: Path to image file or None if not found """ # Check content dictionary if isinstance(element.content, dict): for key in ['saved_path', 'path', 'image_path']: if key in element.content: return element.content[key] # Check direct attribute if hasattr(element, 'saved_path') and element.saved_path: return element.saved_path # Check metadata if element.metadata and isinstance(element.metadata, dict): if 'path' in element.metadata: return element.metadata['path'] if 'saved_path' in element.metadata: return element.metadata['saved_path'] return None def convert_unified_document_to_ocr_data(self, unified_doc: 'UnifiedDocument') -> Dict: """ Convert UnifiedDocument to OCR data format for PDF generation. This method transforms the UnifiedDocument structure into the legacy OCR data format that the PDF generator expects, supporting both OCR and DIRECT processing tracks. Args: unified_doc: UnifiedDocument object from either processing track Returns: Dictionary in OCR data format with text_regions, images_metadata, layout_data """ text_regions = [] images_metadata = [] layout_elements = [] for page in unified_doc.pages: page_num = page.page_number # 1-based for element in page.elements: # Convert BoundingBox to polygon format [[x,y], [x,y], [x,y], [x,y]] bbox_polygon = [ [element.bbox.x0, element.bbox.y0], # top-left [element.bbox.x1, element.bbox.y0], # top-right [element.bbox.x1, element.bbox.y1], # bottom-right [element.bbox.x0, element.bbox.y1], # bottom-left ] # Handle text elements if element.is_text or element.type in [ ElementType.TEXT, ElementType.TITLE, ElementType.HEADER, ElementType.FOOTER, ElementType.PARAGRAPH, ElementType.CAPTION, ElementType.LIST_ITEM, ElementType.FOOTNOTE, ElementType.REFERENCE ]: text_content = element.get_text() if text_content: text_region = { 'text': text_content, 'bbox': bbox_polygon, 'confidence': element.confidence or 1.0, 'page': page_num, 'element_type': element.type.value # Include element type for styling } # Include style information if available (for Direct track) if hasattr(element, 'style') and element.style: text_region['style'] = element.style text_regions.append(text_region) # Handle table elements elif element.type == ElementType.TABLE: # Convert TableData to HTML for layout_data if isinstance(element.content, TableData): html_content = element.content.to_html() elif isinstance(element.content, dict): html_content = element.content.get('html', str(element.content)) else: html_content = str(element.content) table_element = { 'type': 'table', 'content': html_content, 'bbox': [element.bbox.x0, element.bbox.y0, element.bbox.x1, element.bbox.y1], 'page': page_num - 1, # layout uses 0-based 'element_id': element.element_id # For _use_border_only matching } # Preserve cell_boxes and embedded_images from metadata # These are extracted by PP-StructureV3 and used for accurate table rendering if element.metadata: if 'cell_boxes' in element.metadata: table_element['cell_boxes'] = element.metadata['cell_boxes'] table_element['cell_boxes_source'] = element.metadata.get('cell_boxes_source', 'metadata') if 'embedded_images' in element.metadata: table_element['embedded_images'] = element.metadata['embedded_images'] # Pass through rebuild flag - rebuilt tables should use HTML content if element.metadata.get('was_rebuilt'): table_element['was_rebuilt'] = True logger.debug(f"Table {element.element_id}: marked as rebuilt") layout_elements.append(table_element) # Add bbox to images_metadata for text overlap filtering # (no actual image file, just bbox for filtering) img_metadata = { 'image_path': None, # No fake table image 'bbox': bbox_polygon, 'page': page_num - 1, # 0-based for images_metadata 'type': 'table', 'element_id': element.element_id } # Also copy cell_boxes for quality checking if element.metadata and 'cell_boxes' in element.metadata: img_metadata['cell_boxes'] = element.metadata['cell_boxes'] # Mark if table was rebuilt if element.metadata and element.metadata.get('was_rebuilt'): img_metadata['was_rebuilt'] = True images_metadata.append(img_metadata) # Handle image/visual elements (including stamps/seals) elif element.is_visual or element.type in [ ElementType.IMAGE, ElementType.FIGURE, ElementType.CHART, ElementType.DIAGRAM, ElementType.LOGO, ElementType.STAMP ]: # Get image path using fallback logic image_path = self._get_image_path(element) # Only add if we found a valid path if image_path: images_metadata.append({ 'image_path': image_path, 'bbox': bbox_polygon, 'page': page_num - 1, # 0-based 'type': element.type.value }) logger.debug(f"Found image path: {image_path} for element {element.element_id}") else: logger.warning(f"No image path found for visual element {element.element_id}") # Build page dimensions mapping for multi-page support page_dimensions = {} for page in unified_doc.pages: page_dimensions[page.page_number - 1] = { # 0-based index 'width': page.dimensions.width, 'height': page.dimensions.height } # Build OCR data structure ocr_data = { 'text_regions': text_regions, 'images_metadata': images_metadata, 'layout_data': { 'elements': layout_elements, 'total_elements': len(layout_elements) }, 'total_pages': unified_doc.page_count, 'ocr_dimensions': { 'width': unified_doc.pages[0].dimensions.width if unified_doc.pages else 0, 'height': unified_doc.pages[0].dimensions.height if unified_doc.pages else 0 }, 'page_dimensions': page_dimensions, # Per-page dimensions for multi-page support # Metadata for tracking '_from_unified_document': True, '_processing_track': unified_doc.metadata.processing_track.value } logger.info(f"Converted UnifiedDocument to OCR data: " f"{len(text_regions)} text regions, " f"{len(images_metadata)} images, " f"{len(layout_elements)} layout elements, " f"track={unified_doc.metadata.processing_track.value}") return ocr_data def generate_from_unified_document( self, unified_doc: 'UnifiedDocument', output_path: Path, source_file_path: Optional[Path] = None ) -> bool: """ Generate layout-preserving PDF directly from UnifiedDocument. This method supports both OCR and DIRECT processing tracks, preserving layout and coordinate information from either source. Args: unified_doc: UnifiedDocument object output_path: Path to save generated PDF source_file_path: Optional path to original source file Returns: True if successful, False otherwise """ if not UNIFIED_DOCUMENT_AVAILABLE: logger.error("UnifiedDocument support not available") return False try: # Detect processing track for track-specific rendering processing_track = None if hasattr(unified_doc, 'metadata') and unified_doc.metadata: if hasattr(unified_doc.metadata, 'processing_track'): processing_track = unified_doc.metadata.processing_track elif isinstance(unified_doc.metadata, dict): processing_track = unified_doc.metadata.get('processing_track') # Route to track-specific rendering method # ProcessingTrack is (str, Enum), so comparing with enum value works for both string and enum # HYBRID track uses Direct track rendering (Direct text/tables + OCR images) is_direct_track = (processing_track == ProcessingTrack.DIRECT or processing_track == ProcessingTrack.HYBRID) logger.info(f"Processing track: {processing_track}, using {'Direct' if is_direct_track else 'OCR'} track rendering") if is_direct_track: # Direct track: Rich formatting preservation return self._generate_direct_track_pdf( unified_doc=unified_doc, output_path=output_path, source_file_path=source_file_path ) else: # OCR track: Simplified rendering (backward compatible) return self._generate_ocr_track_pdf( unified_doc=unified_doc, output_path=output_path, source_file_path=source_file_path ) except Exception as e: logger.error(f"Failed to generate PDF from UnifiedDocument: {e}") import traceback traceback.print_exc() return False def _is_element_inside_regions(self, element_bbox, regions_elements, overlap_threshold=0.5) -> bool: """ Check if an element overlaps significantly with any exclusion region (table, image). This prevents duplicate rendering when text overlaps with tables/images. Direct extraction often extracts both the structured element (table/image) AND its text content as separate text blocks. Uses overlap ratio detection instead of strict containment, since text blocks from DirectExtractionEngine may be larger than detected table/image regions (e.g., text block includes heading above table). Args: element_bbox: BBox of the element to check regions_elements: List of region elements (tables, images) to check against overlap_threshold: Minimum overlap percentage to trigger filtering (default 0.5 = 50%) Returns: True if element overlaps ≥50% with any region, False otherwise """ if not element_bbox: return False e_x0, e_y0, e_x1, e_y1 = element_bbox.x0, element_bbox.y0, element_bbox.x1, element_bbox.y1 elem_area = (e_x1 - e_x0) * (e_y1 - e_y0) if elem_area <= 0: return False for region in regions_elements: r_bbox = region.bbox if not r_bbox: continue # Calculate overlap rectangle overlap_x0 = max(e_x0, r_bbox.x0) overlap_y0 = max(e_y0, r_bbox.y0) overlap_x1 = min(e_x1, r_bbox.x1) overlap_y1 = min(e_y1, r_bbox.y1) # Check if there is any overlap if overlap_x0 < overlap_x1 and overlap_y0 < overlap_y1: # Calculate overlap area overlap_area = (overlap_x1 - overlap_x0) * (overlap_y1 - overlap_y0) overlap_ratio = overlap_area / elem_area # If element overlaps more than threshold, filter it out if overlap_ratio >= overlap_threshold: return True return False def _generate_direct_track_pdf( self, unified_doc: 'UnifiedDocument', output_path: Path, source_file_path: Optional[Path] = None ) -> bool: """ Generate PDF with rich formatting preservation for Direct track. This method processes UnifiedDocument directly without converting to legacy OCR format, preserving StyleInfo and applying proper text formatting including line breaks. Args: unified_doc: UnifiedDocument from Direct extraction output_path: Path to save generated PDF source_file_path: Optional path to original source file Returns: True if successful, False otherwise """ try: logger.info("=== Direct Track PDF Generation ===") logger.info(f"Total pages: {len(unified_doc.pages)}") # Set current track for helper methods (may be DIRECT or HYBRID) if hasattr(unified_doc, 'metadata') and unified_doc.metadata: self.current_processing_track = unified_doc.metadata.processing_track else: self.current_processing_track = ProcessingTrack.DIRECT # Get page dimensions from first page (for canvas initialization) if not unified_doc.pages: logger.error("No pages in document") return False first_page = unified_doc.pages[0] page_width = first_page.dimensions.width page_height = first_page.dimensions.height logger.info(f"First page dimensions: {page_width} x {page_height}") # Create PDF canvas with first page dimensions (will be updated per page) from reportlab.pdfgen import canvas pdf_canvas = canvas.Canvas(str(output_path), pagesize=(page_width, page_height)) # Process each page for page_idx, page in enumerate(unified_doc.pages): logger.info(f">>> Processing page {page_idx + 1}/{len(unified_doc.pages)}") # Get current page dimensions current_page_width = page.dimensions.width current_page_height = page.dimensions.height logger.info(f"Page {page_idx + 1} dimensions: {current_page_width} x {current_page_height}") if page_idx > 0: pdf_canvas.showPage() # Set page size for current page pdf_canvas.setPageSize((current_page_width, current_page_height)) # Separate elements by type text_elements = [] table_elements = [] image_elements = [] list_elements = [] # FIX: Collect exclusion regions (tables, images) to prevent duplicate rendering regions_to_avoid = [] # Calculate page area for background detection page_area = current_page_width * current_page_height for element in page.elements: if element.type == ElementType.TABLE: table_elements.append(element) regions_to_avoid.append(element) # Tables are exclusion regions elif element.is_visual or element.type in [ ElementType.IMAGE, ElementType.FIGURE, ElementType.CHART, ElementType.DIAGRAM, ElementType.LOGO, ElementType.STAMP ]: # Skip large vector_graphics charts in Direct track # These are visual decorations (borders, lines, frames) that would cover text # PyMuPDF extracts both vector graphics as images AND text layer separately if element.type == ElementType.CHART and element.bbox: content = element.content is_vector_graphics = ( isinstance(content, dict) and content.get('source') == 'vector_graphics' ) if is_vector_graphics: elem_area = (element.bbox.x1 - element.bbox.x0) * (element.bbox.y1 - element.bbox.y0) coverage_ratio = elem_area / page_area if page_area > 0 else 0 if coverage_ratio > 0.5: logger.info(f"Skipping large vector_graphics chart {element.element_id} " f"(covers {coverage_ratio*100:.1f}% of page) - text provides actual content") continue image_elements.append(element) # Only add real images to exclusion regions, NOT charts/diagrams # Charts often have large bounding boxes that include text labels # which should be rendered as selectable text on top if element.type in [ElementType.IMAGE, ElementType.FIGURE, ElementType.LOGO, ElementType.STAMP]: # Check if this is Direct track (text from PDF text layer, not OCR) is_direct = (self.current_processing_track == ProcessingTrack.DIRECT or self.current_processing_track == ProcessingTrack.HYBRID) if is_direct: # Direct track: text is from PDF text layer, not OCR'd from images # Don't exclude any images - text should be rendered on top # This is critical for Office documents with background images logger.debug(f"Direct track: not excluding {element.element_id} from text regions") continue # OCR track: Skip full-page background images from exclusion regions # Smaller images that might contain OCR'd text should still be excluded if element.bbox: elem_area = (element.bbox.x1 - element.bbox.x0) * (element.bbox.y1 - element.bbox.y0) coverage_ratio = elem_area / page_area if page_area > 0 else 0 # If image covers >70% of page, it's likely a background - don't exclude text if coverage_ratio > 0.7: logger.debug(f"OCR track: skipping background image {element.element_id} from exclusion " f"(covers {coverage_ratio*100:.1f}% of page)") continue regions_to_avoid.append(element) elif element.type == ElementType.LIST_ITEM: list_elements.append(element) elif self._is_list_item_fallback(element): # Fallback detection: Check metadata and text patterns list_elements.append(element) # Mark as list item for downstream processing element.type = ElementType.LIST_ITEM elif element.is_text or element.type in [ ElementType.TEXT, ElementType.TITLE, ElementType.HEADER, ElementType.FOOTER, ElementType.PARAGRAPH, ElementType.FOOTNOTE, ElementType.REFERENCE, ElementType.EQUATION, ElementType.CAPTION ]: text_elements.append(element) logger.info(f"Page {page_idx + 1}: {len(text_elements)} text, " f"{len(table_elements)} tables, {len(image_elements)} images, " f"{len(list_elements)} list items") # Use original element order from extraction engine # The extraction engine has already sorted elements by reading order, # handling multi-column layouts correctly (top-to-bottom, left-to-right) all_elements = [] # Preserve original order by iterating through page.elements for elem in page.elements: if elem in image_elements: all_elements.append(('image', elem)) elif elem in table_elements: all_elements.append(('table', elem)) elif elem in list_elements: all_elements.append(('list', elem)) elif elem in text_elements: all_elements.append(('text', elem)) logger.debug(f"Drawing {len(all_elements)} elements in extraction order (preserves multi-column reading order)") logger.debug(f"Exclusion regions: {len(regions_to_avoid)} (tables/images/charts)") # Debug: Log exclusion region types region_types = {} for region in regions_to_avoid: region_type = region.type.name region_types[region_type] = region_types.get(region_type, 0) + 1 if region_types: logger.debug(f" Exclusion region breakdown: {region_types}") # Draw elements in document order for elem_type, elem in all_elements: if elem_type == 'image': self._draw_image_element_direct(pdf_canvas, elem, current_page_height, output_path.parent) elif elem_type == 'table': self._draw_table_element_direct(pdf_canvas, elem, current_page_height) elif elem_type == 'list': # FIX: Check if list item overlaps with table/image if not self._is_element_inside_regions(elem.bbox, regions_to_avoid): self._draw_text_element_direct(pdf_canvas, elem, current_page_height) else: logger.debug(f"Skipping list element {elem.element_id} inside table/image region") elif elem_type == 'text': # FIX: Check if text overlaps with table/image before drawing if not self._is_element_inside_regions(elem.bbox, regions_to_avoid): self._draw_text_element_direct(pdf_canvas, elem, current_page_height) else: logger.debug(f"Skipping text element {elem.element_id} inside table/image region") # Save PDF pdf_canvas.save() logger.info(f"Direct track PDF saved to {output_path}") # Reset track self.current_processing_track = None return True except Exception as e: logger.error(f"Failed to generate Direct track PDF: {e}") import traceback traceback.print_exc() self.current_processing_track = None return False def _generate_ocr_track_pdf( self, unified_doc: 'UnifiedDocument', output_path: Path, source_file_path: Optional[Path] = None ) -> bool: """ Generate PDF with simplified rendering for OCR track. This method uses the existing OCR data conversion and rendering pipeline for backward compatibility. Args: unified_doc: UnifiedDocument from OCR processing output_path: Path to save generated PDF source_file_path: Optional path to original source file Returns: True if successful, False otherwise """ try: logger.info("=== OCR Track PDF Generation ===") # Set current track self.current_processing_track = 'ocr' # Check if simple text positioning mode is enabled if (settings.simple_text_positioning_enabled and TEXT_REGION_RENDERER_AVAILABLE): logger.info("Using simple text positioning mode") result = self._generate_simple_text_pdf( unified_doc=unified_doc, output_path=output_path, source_file_path=source_file_path ) else: # Convert UnifiedDocument to OCR data format (legacy) ocr_data = self.convert_unified_document_to_ocr_data(unified_doc) # Use existing generation pipeline result = self._generate_pdf_from_data( ocr_data=ocr_data, output_path=output_path, source_file_path=source_file_path ) # Reset track self.current_processing_track = None return result except Exception as e: logger.error(f"Failed to generate OCR track PDF: {e}") import traceback traceback.print_exc() self.current_processing_track = None return False def _generate_simple_text_pdf( self, unified_doc: 'UnifiedDocument', output_path: Path, source_file_path: Optional[Path] = None ) -> bool: """ Generate PDF using simple text positioning from raw OCR regions. This approach bypasses complex table structure reconstruction and renders raw OCR text directly at detected positions with rotation correction. Images, charts, figures, seals, and formulas are still rendered normally. Args: unified_doc: UnifiedDocument from OCR processing output_path: Path to save generated PDF source_file_path: Optional path to original source file Returns: True if successful, False otherwise """ try: logger.info("=== Simple Text Positioning PDF Generation ===") # Initialize text region renderer text_renderer = TextRegionRenderer( font_name=self.font_name, debug=settings.simple_text_positioning_debug ) # Get result directory from output_path result_dir = output_path.parent # Try to determine task_id from result directory or output filename # Output path is typically: result_dir/task_id_edited.pdf task_id = None if output_path.stem.endswith('_edited'): task_id = output_path.stem.replace('_edited', '') elif result_dir.name: # result_dir is typically the task_id directory task_id = result_dir.name if not task_id: logger.warning("Could not determine task_id, falling back to legacy method") ocr_data = self.convert_unified_document_to_ocr_data(unified_doc) return self._generate_pdf_from_data( ocr_data=ocr_data, output_path=output_path, source_file_path=source_file_path ) logger.info(f"Task ID: {task_id}, Result dir: {result_dir}") # Get total pages from UnifiedDocument total_pages = len(unified_doc.pages) if unified_doc.pages else 1 # Get page dimensions from first page (for canvas initialization) if not unified_doc.pages: logger.error("No pages in document") return False first_page = unified_doc.pages[0] if hasattr(first_page, 'dimensions') and first_page.dimensions: page_width = float(first_page.dimensions.width) page_height = float(first_page.dimensions.height) else: # Fallback to default size page_width = 612.0 # Letter width page_height = 792.0 # Letter height logger.warning(f"No page dimensions found, using default {page_width}x{page_height}") logger.info(f"Initial page size: {page_width:.1f} x {page_height:.1f}") # Create PDF canvas pdf_canvas = canvas.Canvas(str(output_path), pagesize=(page_width, page_height)) # Collect image-type elements from UnifiedDocument for rendering # Types that should be rendered as images: figure, image, chart, seal, formula image_element_types = {'figure', 'image', 'chart', 'seal', 'formula'} # Process each page for page_num in range(1, total_pages + 1): logger.info(f">>> Processing page {page_num}/{total_pages}") # Get page dimensions for current page if page_num <= len(unified_doc.pages): current_page = unified_doc.pages[page_num - 1] if hasattr(current_page, 'dimensions') and current_page.dimensions: current_width = float(current_page.dimensions.width) current_height = float(current_page.dimensions.height) else: current_width = page_width current_height = page_height else: current_width = page_width current_height = page_height if page_num > 1: pdf_canvas.showPage() # Set page size pdf_canvas.setPageSize((current_width, current_height)) # === Layer 1: Render images, charts, figures, seals, formulas === # Also collect exclusion zones for text avoidance exclusion_zones = [] # List of (x0, y0, x1, y1) tuples if page_num <= len(unified_doc.pages): current_page = unified_doc.pages[page_num - 1] page_elements = current_page.elements if hasattr(current_page, 'elements') else [] image_elements_rendered = 0 for elem in page_elements: elem_type = elem.type if hasattr(elem, 'type') else elem.get('type', '') # Handle enum type if hasattr(elem_type, 'value'): elem_type = elem_type.value if elem_type in image_element_types: # Get image path from element content content = elem.content if hasattr(elem, 'content') else elem.get('content', {}) if isinstance(content, dict): saved_path = content.get('saved_path') or content.get('path') else: saved_path = None # Get bbox for exclusion zone (even if image file not found) bbox = elem.bbox if hasattr(elem, 'bbox') else elem.get('bbox', {}) if hasattr(bbox, 'x0'): x0, y0, x1, y1 = bbox.x0, bbox.y0, bbox.x1, bbox.y1 elif isinstance(bbox, dict): x0 = bbox.get('x0', 0) y0 = bbox.get('y0', 0) x1 = bbox.get('x1', x0 + bbox.get('width', 0)) y1 = bbox.get('y1', y0 + bbox.get('height', 0)) else: continue # Add to exclusion zones for text avoidance # Use original image coordinates (not PDF flipped) exclusion_zones.append((x0, y0, x1, y1)) if saved_path: # Try to find the image file image_path = result_dir / saved_path if not image_path.exists(): # Try in imgs subdirectory image_path = result_dir / 'imgs' / saved_path if not image_path.exists(): # Try just the filename image_path = result_dir / Path(saved_path).name if image_path.exists(): try: # Convert coordinates (flip Y for PDF) pdf_x = x0 pdf_y = current_height - y1 # Bottom of image in PDF coords img_width = x1 - x0 img_height = y1 - y0 # Draw image pdf_canvas.drawImage( str(image_path), pdf_x, pdf_y, width=img_width, height=img_height, preserveAspectRatio=True, mask='auto' ) image_elements_rendered += 1 logger.debug(f"Rendered {elem_type}: {saved_path} at ({pdf_x:.1f}, {pdf_y:.1f})") except Exception as e: logger.warning(f"Failed to render {elem_type} {saved_path}: {e}") else: logger.warning(f"Image file not found: {saved_path}") if image_elements_rendered > 0: logger.info(f"Rendered {image_elements_rendered} image elements (figures/charts/seals/formulas)") if exclusion_zones: logger.info(f"Collected {len(exclusion_zones)} exclusion zones for text avoidance") # === Layer 2: Render text from raw OCR regions === raw_regions = load_raw_ocr_regions(str(result_dir), task_id, page_num) if not raw_regions: logger.warning(f"No raw OCR regions found for page {page_num}") else: logger.info(f"Loaded {len(raw_regions)} raw OCR regions for page {page_num}") # Collect texts inside exclusion zones for position-aware deduplication # This prevents duplicate axis labels from being rendered near charts zone_texts = None if exclusion_zones: zone_texts = text_renderer.collect_zone_texts( raw_regions, exclusion_zones, threshold=0.5, include_axis_labels=True ) if zone_texts: logger.info(f"Collected {len(zone_texts)} zone texts for deduplication: {list(zone_texts)[:10]}...") # Render all text regions, avoiding exclusion zones (images/charts) # Scale factors are 1.0 since OCR dimensions match page dimensions rendered = text_renderer.render_all_regions( pdf_canvas=pdf_canvas, regions=raw_regions, page_height=current_height, scale_x=1.0, scale_y=1.0, exclusion_zones=exclusion_zones, zone_texts=zone_texts ) logger.info(f"Rendered {rendered} text regions") logger.info(f"<<< Page {page_num} complete") # Save PDF pdf_canvas.save() file_size = output_path.stat().st_size logger.info(f"Generated PDF: {output_path.name} ({file_size} bytes)") return True except Exception as e: logger.error(f"Failed to generate simple text PDF: {e}") import traceback traceback.print_exc() return False def _generate_pdf_from_data( self, ocr_data: Dict, output_path: Path, source_file_path: Optional[Path] = None, json_parent_dir: Optional[Path] = None ) -> bool: """ Internal method to generate PDF from OCR data dictionary. This is the core generation logic extracted for reuse by both JSON-based and UnifiedDocument-based generation paths. Args: ocr_data: OCR data dictionary output_path: Path to save generated PDF source_file_path: Optional path to original source file json_parent_dir: Directory containing images (for JSON-based generation) Returns: True if successful, False otherwise """ try: # Note: Removed PDF caching - always regenerate to ensure latest code changes take effect # If caching is needed, implement at a higher level with proper cache invalidation # Get text regions text_regions = ocr_data.get('text_regions', []) if not text_regions: logger.warning("No text regions found in data") # Don't fail - might have only tables/images # Get images metadata images_metadata = ocr_data.get('images_metadata', []) # Get layout data layout_data = ocr_data.get('layout_data', {}) # Step 1: Get OCR processing dimensions (for first page / default) ocr_width, ocr_height = self.calculate_page_dimensions(ocr_data, source_file_path=None) logger.info(f"OCR 處理時使用的座標系尺寸 (第一頁): {ocr_width:.1f} x {ocr_height:.1f}") # Step 2: Get page dimensions mapping for multi-page support page_dimensions = ocr_data.get('page_dimensions', {}) if not page_dimensions: # Fallback: use first page dimensions for all pages page_dimensions = {0: {'width': ocr_width, 'height': ocr_height}} logger.info("No page_dimensions found, using first page size for all pages") # Step 3: Get original file dimensions for all pages # For OCR track, we use OCR coordinate system dimensions directly to avoid scaling issues original_page_sizes = {} use_ocr_dimensions_for_pdf = (self.current_processing_track == 'ocr') if use_ocr_dimensions_for_pdf: # OCR Track: Use OCR coordinate system dimensions directly # This ensures no scaling is needed (scale = 1.0) logger.info(f"OCR Track: 使用 OCR 座標系尺寸作為 PDF 頁面尺寸（避免縮放）") elif source_file_path: original_page_sizes = self.get_all_page_sizes(source_file_path) if original_page_sizes: logger.info(f"從原始文件獲取到 {len(original_page_sizes)} 頁尺寸") else: logger.warning(f"無法獲取原始文件尺寸，將使用 OCR/UnifiedDocument 尺寸") else: logger.info(f"無原始文件，將使用 OCR/UnifiedDocument 尺寸") # Determine initial canvas size (will be updated per page) # Priority for OCR track: OCR dimensions (no scaling) # Priority for Direct track: original file first page > OCR/UnifiedDocument first page if use_ocr_dimensions_for_pdf: target_width, target_height = ocr_width, ocr_height logger.info(f"初始 PDF 尺寸（OCR Track, 使用 OCR 座標系）: {target_width:.1f} x {target_height:.1f}") elif 0 in original_page_sizes: target_width, target_height = original_page_sizes[0] logger.info(f"初始 PDF 尺寸（來自原始文件首頁）: {target_width:.1f} x {target_height:.1f}") else: target_width, target_height = ocr_width, ocr_height logger.info(f"初始 PDF 尺寸（來自 OCR/UnifiedDocument）: {target_width:.1f} x {target_height:.1f}") # Step 4: Detect content orientation mismatch # This handles rotated scans where content bbox exceeds page dimensions # IMPORTANT: Use OCR dimensions (pixels) for detection, not PDF points # because content bboxes are in the same coordinate system as OCR dimensions needs_rotation, adjusted_ocr_width, adjusted_ocr_height = self._detect_content_orientation( ocr_width, ocr_height, ocr_data ) # If orientation change detected, calculate the adjusted target dimensions if needs_rotation: # Swap target dimensions to match the detected orientation adjusted_width = target_height adjusted_height = target_width elif adjusted_ocr_width != ocr_width or adjusted_ocr_height != ocr_height: # Content extends beyond OCR dimensions, scale target proportionally scale_w = adjusted_ocr_width / ocr_width if ocr_width > 0 else 1.0 scale_h = adjusted_ocr_height / ocr_height if ocr_height > 0 else 1.0 adjusted_width = target_width * scale_w adjusted_height = target_height * scale_h else: adjusted_width = target_width adjusted_height = target_height if needs_rotation or (adjusted_width != target_width or adjusted_height != target_height): logger.info(f"頁面尺寸調整: {target_width:.1f}x{target_height:.1f} -> " f"{adjusted_width:.1f}x{adjusted_height:.1f} (旋轉={needs_rotation})") target_width, target_height = adjusted_width, adjusted_height # Update original_page_sizes with the new TARGET dimensions if 0 in original_page_sizes: original_page_sizes[0] = (target_width, target_height) logger.info(f"覆蓋原始文件尺寸以適應內容方向") # CRITICAL: Update page_dimensions with SWAPPED OCR dimensions # This is the coordinate system that the content bboxes are in # When content is rotated, width and height are effectively swapped if needs_rotation and 0 in page_dimensions: # Swap the OCR dimensions to match the rotated content coordinate system original_ocr_w = page_dimensions[0]['width'] original_ocr_h = page_dimensions[0]['height'] page_dimensions[0] = {'width': original_ocr_h, 'height': original_ocr_w} logger.info(f"旋轉 OCR 座標系: {original_ocr_w:.1f}x{original_ocr_h:.1f} -> " f"{original_ocr_h:.1f}x{original_ocr_w:.1f}") # Create PDF canvas with initial page size (will be updated per page) pdf_canvas = canvas.Canvas(str(output_path), pagesize=(target_width, target_height)) # Smart filtering: only include tables with good cell_boxes quality in regions_to_avoid # Tables with bad cell_boxes will use raw OCR text positioning instead # Exception: Rebuilt tables always use HTML content and filter text regions_to_avoid = [] good_quality_tables = [] bad_quality_tables = [] rebuilt_tables = [] for img in images_metadata: if img.get('type') == 'table': elem_id = img.get('element_id', 'unknown') # Check if this table was rebuilt - rebuilt tables have good content was_rebuilt = img.get('was_rebuilt', False) if was_rebuilt: # Rebuilt tables have accurate content - filter text, use HTML regions_to_avoid.append(img) rebuilt_tables.append(elem_id) else: # Check cell_boxes quality for non-rebuilt tables cell_boxes = img.get('cell_boxes', []) quality = self._check_cell_boxes_quality(cell_boxes, elem_id) if quality == 'good': # Good quality: filter text, render with cell_boxes regions_to_avoid.append(img) good_quality_tables.append(elem_id) else: # Bad quality: don't filter text, just draw border bad_quality_tables.append(elem_id) img['_use_border_only'] = True # Mark for border-only rendering else: # Non-table elements (images, figures, charts) always avoid regions_to_avoid.append(img) logger.info(f"過濾文字區域: {len(regions_to_avoid)} 個區域需要避免") if rebuilt_tables: logger.info(f" 重建表格用 HTML: {rebuilt_tables}") if good_quality_tables: logger.info(f" 表格用 cell_boxes: {good_quality_tables}") if bad_quality_tables: logger.info(f" 表格用 raw OCR text (border only): {bad_quality_tables}") filtered_text_regions = self._filter_text_in_regions(text_regions, regions_to_avoid) # Group regions by page pages_data = {} for region in filtered_text_regions: page_num = region.get('page', 1) if page_num not in pages_data: pages_data[page_num] = [] pages_data[page_num].append(region) # Get table elements from layout_data and copy _use_border_only flags table_elements = [] if layout_data and layout_data.get('elements'): # Create a lookup for _use_border_only flags from images_metadata border_only_tables = {img.get('element_id') for img in images_metadata if img.get('type') == 'table' and img.get('_use_border_only')} logger.debug(f"[DEBUG] border_only_tables from images_metadata: {border_only_tables}") for e in layout_data['elements']: if e.get('type') == 'table': elem_id = e.get('element_id') logger.debug(f"[DEBUG] layout_data table element_id: {elem_id}") # Copy the flag if this table should use border only if elem_id in border_only_tables: e['_use_border_only'] = True logger.info(f"[DEBUG] Set _use_border_only=True for table {elem_id}") table_elements.append(e) # Process each page total_pages = ocr_data.get('total_pages', 1) logger.info(f"開始處理 {total_pages} 頁 PDF") # Determine image directory if json_parent_dir is None: json_parent_dir = output_path.parent for page_num in range(1, total_pages + 1): logger.info(f">>> 處理第 {page_num}/{total_pages} 頁") # Get current page dimensions with priority order: # For OCR Track: always use OCR dimensions (scale = 1.0) # For Direct Track: # 1. Original file dimensions (highest priority) # 2. OCR/UnifiedDocument dimensions # 3. Fallback to first page dimensions page_idx = page_num - 1 dimension_source = "unknown" # For OCR Track: always use OCR dimensions if use_ocr_dimensions_for_pdf and page_idx in page_dimensions: current_page_dims = page_dimensions[page_idx] current_target_w = float(current_page_dims['width']) current_target_h = float(current_page_dims['height']) dimension_source = "ocr_track_direct" # Priority 1: Original file dimensions (Direct Track only) elif page_idx in original_page_sizes: current_target_w, current_target_h = original_page_sizes[page_idx] dimension_source = "original_file" # Priority 2: OCR/UnifiedDocument dimensions (which may have been adjusted for orientation) elif page_idx in page_dimensions: current_page_dims = page_dimensions[page_idx] current_target_w = float(current_page_dims['width']) current_target_h = float(current_page_dims['height']) dimension_source = "ocr_unified_doc" # Priority 3: Fallback to first page else: current_target_w = ocr_width current_target_h = ocr_height dimension_source = "fallback_first_page" logger.warning(f"No dimensions for page {page_num}, using first page size") # For pages after the first, check if orientation adjustment is needed # (First page was already handled above) if page_num > 1 and dimension_source == "original_file": # Build per-page data for orientation detection page_ocr_data = { 'text_regions': [r for r in text_regions if r.get('page', 1) == page_num], 'layout_data': { 'elements': [e for e in layout_data.get('elements', []) if e.get('page', 0) == page_idx] }, 'images_metadata': [i for i in images_metadata if i.get('page', 0) == page_idx] } needs_page_rotation, adj_w, adj_h = self._detect_content_orientation( current_target_w, current_target_h, page_ocr_data ) if needs_page_rotation or (adj_w != current_target_w or adj_h != current_target_h): logger.info(f"第 {page_num} 頁尺寸調整: " f"{current_target_w:.1f}x{current_target_h:.1f} -> " f"{adj_w:.1f}x{adj_h:.1f}") current_target_w, current_target_h = adj_w, adj_h # Calculate scale factors for coordinate transformation # OCR coordinates need to be scaled if original file dimensions differ if dimension_source == "original_file": # Get OCR dimensions for this page to calculate scale if page_idx in page_dimensions: ocr_page_w = float(page_dimensions[page_idx]['width']) ocr_page_h = float(page_dimensions[page_idx]['height']) else: ocr_page_w = ocr_width ocr_page_h = ocr_height current_scale_w = current_target_w / ocr_page_w if ocr_page_w > 0 else 1.0 current_scale_h = current_target_h / ocr_page_h if ocr_page_h > 0 else 1.0 else: # Using OCR/UnifiedDocument dimensions directly, no scaling needed current_scale_w = 1.0 current_scale_h = 1.0 logger.info(f"第 {page_num} 頁尺寸: {current_target_w:.1f} x {current_target_h:.1f} " f"(來源: {dimension_source}, 縮放: {current_scale_w:.3f}x{current_scale_h:.3f})") if page_num > 1: pdf_canvas.showPage() # Set page size for current page pdf_canvas.setPageSize((current_target_w, current_target_h)) # Get regions for this page page_text_regions = pages_data.get(page_num, []) page_table_regions = [t for t in table_elements if t.get('page') == page_num - 1] page_image_regions = [ img for img in images_metadata if img.get('page') == page_num - 1 and img.get('type') != 'table' and img.get('image_path') is not None # Skip table placeholders ] # Draw in layers: images → tables → text # 1. Draw images (bottom layer) for img_meta in page_image_regions: self.draw_image_region( pdf_canvas, img_meta, current_target_h, json_parent_dir, current_scale_w, current_scale_h ) # 2. Draw tables (middle layer) for table_elem in page_table_regions: self.draw_table_region( pdf_canvas, table_elem, images_metadata, current_target_h, current_scale_w, current_scale_h, result_dir=json_parent_dir ) # 3. Draw text (top layer) for region in page_text_regions: self.draw_text_region( pdf_canvas, region, current_target_h, current_scale_w, current_scale_h ) logger.info(f"<<< 第 {page_num} 頁完成") # Save PDF pdf_canvas.save() file_size = output_path.stat().st_size logger.info(f"Generated PDF: {output_path.name} ({file_size} bytes)") return True except Exception as e: logger.error(f"Failed to generate PDF: {e}") import traceback traceback.print_exc() return False def calculate_page_dimensions(self, ocr_data: Dict, source_file_path: Optional[Path] = None) -> Tuple[float, float]: """ 從 OCR JSON 數據中取得頁面尺寸。 優先使用明確的 dimensions 欄位，失敗時才回退到 bbox 推斷。 Args: ocr_data: Complete OCR data dictionary with text_regions and layout source_file_path: Optional path to source file (fallback only) Returns: Tuple of (width, height) in points """ # *** 優先級 1: 檢查 ocr_dimensions (UnifiedDocument 轉換來的) *** if 'ocr_dimensions' in ocr_data: dims = ocr_data['ocr_dimensions'] # Handle both dict format {'width': w, 'height': h} and # list format [{'page': 1, 'width': w, 'height': h}, ...] if isinstance(dims, list) and len(dims) > 0: dims = dims[0] # Use first page dimensions if isinstance(dims, dict): w = float(dims.get('width', 0)) h = float(dims.get('height', 0)) if w > 0 and h > 0: logger.info(f"使用 ocr_dimensions 欄位的頁面尺寸: {w:.1f} x {h:.1f}") return (w, h) # *** 優先級 2: 檢查原始 JSON 的 dimensions *** if 'dimensions' in ocr_data: dims = ocr_data['dimensions'] w = float(dims.get('width', 0)) h = float(dims.get('height', 0)) if w > 0 and h > 0: logger.info(f"使用 dimensions 欄位的頁面尺寸: {w:.1f} x {h:.1f}") return (w, h) # *** 優先級 3: Fallback - 從 bbox 推斷 (僅當上述皆缺失時使用) *** logger.info("dimensions 欄位不可用，回退到 bbox 推斷") max_x = 0 max_y = 0 # *** 關鍵修復：檢查所有可能包含 bbox 的字段 *** # 不同版本的 OCR 輸出可能使用不同的字段名 all_regions = [] # 1. text_regions - 包含所有文字區域（最常見） if 'text_regions' in ocr_data and isinstance(ocr_data['text_regions'], list): all_regions.extend(ocr_data['text_regions']) # 2. image_regions - 包含圖片區域 if 'image_regions' in ocr_data and isinstance(ocr_data['image_regions'], list): all_regions.extend(ocr_data['image_regions']) # 3. tables - 包含表格區域 if 'tables' in ocr_data and isinstance(ocr_data['tables'], list): all_regions.extend(ocr_data['tables']) # 4. layout - 可能包含布局信息（可能是空列表） if 'layout' in ocr_data and isinstance(ocr_data['layout'], list): all_regions.extend(ocr_data['layout']) # 5. layout_data.elements - PP-StructureV3 格式 if 'layout_data' in ocr_data and isinstance(ocr_data['layout_data'], dict): elements = ocr_data['layout_data'].get('elements', []) if elements: all_regions.extend(elements) if not all_regions: # 如果 JSON 為空，回退到原始檔案尺寸 logger.warning("JSON 中沒有找到 text_regions, image_regions, tables, layout 或 layout_data.elements，回退到原始檔案尺寸。") if source_file_path: dims = self.get_original_page_size(source_file_path) if dims: return dims return A4 region_count = 0 for region in all_regions: try: bbox = region.get('bbox') if not bbox: continue region_count += 1 # *** 關鍵修復：正確處理多邊形 [[x, y], ...] 格式 *** if isinstance(bbox[0], (int, float)): # 處理簡單的 [x1, y1, x2, y2] 格式 max_x = max(max_x, bbox[2]) max_y = max(max_y, bbox[3]) elif isinstance(bbox[0], (list, tuple)): # 處理多邊形 [[x, y], ...] 格式 x_coords = [p[0] for p in bbox if isinstance(p, (list, tuple)) and len(p) >= 2] y_coords = [p[1] for p in bbox if isinstance(p, (list, tuple)) and len(p) >= 2] if x_coords and y_coords: max_x = max(max_x, max(x_coords)) max_y = max(max_y, max(y_coords)) except Exception as e: logger.warning(f"Error processing bbox {bbox}: {e}") if max_x > 0 and max_y > 0: logger.info(f"從 {region_count} 個區域中推斷出的 OCR 座標系尺寸: {max_x:.1f} x {max_y:.1f}") return (max_x, max_y) else: # 如果所有 bbox 都解析失敗，才回退 logger.warning("無法從 bbox 推斷尺寸，回退到原始檔案尺寸。") if source_file_path: dims = self.get_original_page_size(source_file_path) if dims: return dims return A4 def get_all_page_sizes(self, file_path: Path) -> Dict[int, Tuple[float, float]]: """ Extract dimensions for all pages from original source file Args: file_path: Path to original file (image or PDF) Returns: Dict mapping page index (0-based) to (width, height) in points Empty dict if extraction fails """ page_sizes = {} try: if not file_path.exists(): logger.warning(f"File not found: {file_path}") return page_sizes # For images, single page with dimensions from PIL if file_path.suffix.lower() in ['.png', '.jpg', '.jpeg', '.bmp', '.tiff']: img = Image.open(file_path) # Use pixel dimensions directly as points (1:1 mapping) # This matches how PaddleOCR reports coordinates width_pt = float(img.width) height_pt = float(img.height) page_sizes[0] = (width_pt, height_pt) logger.info(f"Extracted dimensions from image: {width_pt:.1f} x {height_pt:.1f} points (1:1 pixel mapping)") return page_sizes # For PDFs, extract dimensions for all pages using PyPDF2 if file_path.suffix.lower() == '.pdf': try: from PyPDF2 import PdfReader reader = PdfReader(file_path) total_pages = len(reader.pages) for page_idx in range(total_pages): page = reader.pages[page_idx] # MediaBox gives [x1, y1, x2, y2] in points mediabox = page.mediabox width_pt = float(mediabox.width) height_pt = float(mediabox.height) # IMPORTANT: Consider page rotation! # PDF pages can have /Rotate attribute (0, 90, 180, 270) # When rotation is 90 or 270 degrees, width and height should be swapped # because pdf2image and PDF viewers apply this rotation when rendering rotation = page.get('/Rotate', 0) if rotation is None: rotation = 0 rotation = int(rotation) % 360 if rotation in (90, 270): # Swap width and height for 90/270 degree rotation width_pt, height_pt = height_pt, width_pt logger.info(f"Page {page_idx}: Rotation={rotation}°, swapped dimensions to {width_pt:.1f} x {height_pt:.1f}") page_sizes[page_idx] = (width_pt, height_pt) logger.info(f"Extracted dimensions from PDF: {total_pages} pages") for idx, (w, h) in page_sizes.items(): logger.debug(f" Page {idx}: {w:.1f} x {h:.1f} points") return page_sizes except ImportError: logger.warning("PyPDF2 not available, cannot extract PDF dimensions") except Exception as e: logger.warning(f"Failed to extract PDF dimensions: {e}") except Exception as e: logger.warning(f"Failed to get page sizes from {file_path}: {e}") return page_sizes def get_original_page_size(self, file_path: Path) -> Optional[Tuple[float, float]]: """ Extract first page dimensions from original source file (backward compatibility) Args: file_path: Path to original file (image or PDF) Returns: Tuple of (width, height) in points or None """ page_sizes = self.get_all_page_sizes(file_path) if 0 in page_sizes: return page_sizes[0] return None def _get_bbox_coords(self, bbox: Union[Dict, List[List[float]], List[float]]) -> Optional[Tuple[float, float, float, float]]: """將任何 bbox 格式 (dict, 多邊形或 [x1,y1,x2,y2]) 轉換為 [x_min, y_min, x_max, y_max]""" try: if bbox is None: return None # Dict format from UnifiedDocument: {"x0": ..., "y0": ..., "x1": ..., "y1": ...} if isinstance(bbox, dict): if 'x0' in bbox and 'y0' in bbox and 'x1' in bbox and 'y1' in bbox: return float(bbox['x0']), float(bbox['y0']), float(bbox['x1']), float(bbox['y1']) else: logger.warning(f"Dict bbox 缺少必要欄位: {bbox}") return None if not isinstance(bbox, (list, tuple)) or len(bbox) < 4: return None if isinstance(bbox[0], (list, tuple)): # 處理多邊形 [[x, y], ...] x_coords = [p[0] for p in bbox if isinstance(p, (list, tuple)) and len(p) >= 2] y_coords = [p[1] for p in bbox if isinstance(p, (list, tuple)) and len(p) >= 2] if not x_coords or not y_coords: return None return min(x_coords), min(y_coords), max(x_coords), max(y_coords) elif isinstance(bbox[0], (int, float)) and len(bbox) == 4: # 處理 [x1, y1, x2, y2] return float(bbox[0]), float(bbox[1]), float(bbox[2]), float(bbox[3]) else: logger.warning(f"未知的 bbox 格式: {bbox}") return None except Exception as e: logger.error(f"解析 bbox {bbox} 時出錯: {e}") return None def _is_bbox_inside(self, inner_bbox_data: Dict, outer_bbox_data: Dict, tolerance: float = 5.0) -> bool: """ 檢查 'inner_bbox' 是否在 'outer_bbox' 內部（帶有容錯）。 此版本可處理多邊形和矩形。 """ inner_coords = self._get_bbox_coords(inner_bbox_data.get('bbox')) outer_coords = self._get_bbox_coords(outer_bbox_data.get('bbox')) if not inner_coords or not outer_coords: return False inner_x1, inner_y1, inner_x2, inner_y2 = inner_coords outer_x1, outer_y1, outer_x2, outer_y2 = outer_coords # 檢查 inner 是否在 outer 內部 (加入 tolerance) is_inside = ( (inner_x1 >= outer_x1 - tolerance) and (inner_y1 >= outer_y1 - tolerance) and (inner_x2 <= outer_x2 + tolerance) and (inner_y2 <= outer_y2 + tolerance) ) return is_inside def _bbox_overlaps(self, bbox1_data: Dict, bbox2_data: Dict, tolerance: float = 5.0) -> bool: """ 檢查兩個 bbox 是否有重疊（帶有容錯）。 如果有任何重疊，返回 True。 Args: bbox1_data: 第一個 bbox 數據 bbox2_data: 第二個 bbox 數據 tolerance: 容錯值（像素） Returns: True 如果兩個 bbox 有重疊 """ coords1 = self._get_bbox_coords(bbox1_data.get('bbox')) coords2 = self._get_bbox_coords(bbox2_data.get('bbox')) if not coords1 or not coords2: return False x1_min, y1_min, x1_max, y1_max = coords1 x2_min, y2_min, x2_max, y2_max = coords2 # 擴展 bbox2（表格/圖片區域）的範圍 x2_min -= tolerance y2_min -= tolerance x2_max += tolerance y2_max += tolerance # 檢查是否有重疊：如果沒有重疊，則必定滿足以下條件之一 no_overlap = ( x1_max < x2_min or # bbox1 在 bbox2 左側 x1_min > x2_max or # bbox1 在 bbox2 右側 y1_max < y2_min or # bbox1 在 bbox2 上方 y1_min > y2_max # bbox1 在 bbox2 下方 ) return not no_overlap def _calculate_overlap_ratio(self, text_bbox_data: Dict, avoid_bbox_data: Dict) -> float: """ 計算文字區域與避免區域的重疊比例。 Args: text_bbox_data: 文字區域 bbox 數據 avoid_bbox_data: 避免區域 bbox 數據 Returns: 重疊面積佔文字區域面積的比例 (0.0 - 1.0) """ text_coords = self._get_bbox_coords(text_bbox_data.get('bbox')) avoid_coords = self._get_bbox_coords(avoid_bbox_data.get('bbox')) if not text_coords or not avoid_coords: return 0.0 tx0, ty0, tx1, ty1 = text_coords ax0, ay0, ax1, ay1 = avoid_coords # Calculate text area text_area = (tx1 - tx0) * (ty1 - ty0) if text_area <= 0: return 0.0 # Calculate intersection inter_x0 = max(tx0, ax0) inter_y0 = max(ty0, ay0) inter_x1 = min(tx1, ax1) inter_y1 = min(ty1, ay1) # Check if there's actual intersection if inter_x1 <= inter_x0 or inter_y1 <= inter_y0: return 0.0 inter_area = (inter_x1 - inter_x0) * (inter_y1 - inter_y0) return inter_area / text_area def _filter_text_in_regions(self, text_regions: List[Dict], regions_to_avoid: List[Dict], overlap_threshold: float = 0.5) -> List[Dict]: """ 過濾掉與 'regions_to_avoid'（例如表格、圖片）顯著重疊的文字區域。 使用重疊比例閾值來判斷是否過濾，避免過濾掉僅相鄰但不重疊的文字。 Args: text_regions: 文字區域列表 regions_to_avoid: 需要避免的區域列表（表格、圖片） overlap_threshold: 重疊比例閾值 (0.0-1.0)，只有當文字區域 與避免區域的重疊比例超過此閾值時才會被過濾 預設 0.5 表示超過 50% 重疊才過濾 Returns: 過濾後的文字區域列表 """ filtered_text = [] filtered_count = 0 for text_region in text_regions: should_filter = False max_overlap = 0.0 for avoid_region in regions_to_avoid: # 計算重疊比例 overlap_ratio = self._calculate_overlap_ratio(text_region, avoid_region) max_overlap = max(max_overlap, overlap_ratio) # 只有當重疊比例超過閾值時才過濾 if overlap_ratio > overlap_threshold: should_filter = True filtered_count += 1 logger.debug(f"過濾掉重疊文字 (重疊比例: {overlap_ratio:.1%}): {text_region.get('text', '')[:30]}...") break if not should_filter: filtered_text.append(text_region) if max_overlap > 0: logger.debug(f"保留文字 (最大重疊比例: {max_overlap:.1%}): {text_region.get('text', '')[:30]}...") logger.info(f"原始文字區域: {len(text_regions)}, 過濾後: {len(filtered_text)}, 移除: {filtered_count}") return filtered_text def draw_text_region( self, pdf_canvas: canvas.Canvas, region: Dict, page_height: float, scale_w: float = 1.0, scale_h: float = 1.0 ): """ Draw a text region at precise coordinates Args: pdf_canvas: ReportLab canvas object region: Text region dict with text, bbox, confidence page_height: Height of page (for coordinate transformation) scale_w: Scale factor for X coordinates (PDF width / OCR width) scale_h: Scale factor for Y coordinates (PDF height / OCR height) """ text = region.get('text', '') bbox = region.get('bbox', []) confidence = region.get('confidence', 1.0) if not text or not bbox: return try: # Handle different bbox formats if isinstance(bbox, dict): # Dict format from UnifiedDocument: {"x0": ..., "y0": ..., "x1": ..., "y1": ...} if 'x0' in bbox and 'y0' in bbox and 'x1' in bbox and 'y1' in bbox: ocr_x_left = float(bbox['x0']) ocr_y_top = float(bbox['y0']) ocr_x_right = float(bbox['x1']) ocr_y_bottom = float(bbox['y1']) else: logger.warning(f"Dict bbox missing required keys: {bbox}") return elif isinstance(bbox, list): if len(bbox) < 4: return # Polygon format [[x,y], [x,y], [x,y], [x,y]] (4 points) if isinstance(bbox[0], list): ocr_x_left = bbox[0][0] # Left X ocr_y_top = bbox[0][1] # Top Y in OCR coordinates ocr_x_right = bbox[2][0] # Right X ocr_y_bottom = bbox[2][1] # Bottom Y in OCR coordinates # Simple list format [x0, y0, x1, y1] elif isinstance(bbox[0], (int, float)): ocr_x_left = bbox[0] ocr_y_top = bbox[1] ocr_x_right = bbox[2] ocr_y_bottom = bbox[3] else: logger.warning(f"Unexpected bbox list format: {bbox}") return else: logger.warning(f"Invalid bbox format: {bbox}") return logger.info(f"[文字] '{text[:20]}...' OCR原始座標: L={ocr_x_left:.0f}, T={ocr_y_top:.0f}, R={ocr_x_right:.0f}, B={ocr_y_bottom:.0f}") # Apply scale factors to convert from OCR space to PDF space scaled_x_left = ocr_x_left * scale_w scaled_y_top = ocr_y_top * scale_h scaled_x_right = ocr_x_right * scale_w scaled_y_bottom = ocr_y_bottom * scale_h logger.info(f"[文字] '{text[:20]}...' 縮放後(scale={scale_w:.3f},{scale_h:.3f}): L={scaled_x_left:.1f}, T={scaled_y_top:.1f}, R={scaled_x_right:.1f}, B={scaled_y_bottom:.1f}") # Calculate bbox dimensions (after scaling) bbox_width = abs(scaled_x_right - scaled_x_left) bbox_height = abs(scaled_y_bottom - scaled_y_top) # Calculate font size using heuristics # For multi-line text, divide bbox height by number of lines lines = text.split('\n') non_empty_lines = [l for l in lines if l.strip()] num_lines = max(len(non_empty_lines), 1) # Font size calculation with stabilization # Use 0.8 factor to leave room for line spacing raw_font_size = (bbox_height / num_lines) * 0.8 # Stabilize font size for body text (most common case) # Normal body text should be 9-11pt, only deviate for clear outliers element_type = region.get('element_type', 'text') if element_type in ('text', 'paragraph'): # For body text, bias toward 10pt baseline if 7 <= raw_font_size <= 14: # Near-normal range: use weighted average toward 10pt font_size = raw_font_size * 0.7 + 10 * 0.3 else: # Clear outlier: use raw but clamp more aggressively font_size = max(min(raw_font_size, 14), 7) else: # For titles/headers/etc, use raw calculation with wider range font_size = max(min(raw_font_size, 72), 4) logger.debug(f"Text has {num_lines} non-empty lines, bbox_height={bbox_height:.1f}, raw={raw_font_size:.1f}, final={font_size:.1f}") # Transform coordinates: OCR (top-left origin) → PDF (bottom-left origin) # CRITICAL: Y-axis flip! # For multi-line text, start from TOP of bbox and go downward pdf_x = scaled_x_left pdf_y_top = page_height - scaled_y_top # Top of bbox in PDF coordinates # Adjust for font baseline: first line starts below the top edge pdf_y = pdf_y_top - font_size # Start first line one font size below top logger.info(f"[文字] '{text[:30]}' → PDF位置: ({pdf_x:.1f}, {pdf_y:.1f}), 字體:{font_size:.1f}pt, 寬x高:{bbox_width:.0f}x{bbox_height:.0f}, 行數:{num_lines}") # Set font with track-specific styling style_info = region.get('style') element_type = region.get('element_type', 'text') is_direct_track = (self.current_processing_track == ProcessingTrack.DIRECT or self.current_processing_track == ProcessingTrack.HYBRID) if style_info and is_direct_track: # Direct track: Apply rich styling from StyleInfo self._apply_text_style(pdf_canvas, style_info, default_size=font_size) # Get current font for width calculation font_name = pdf_canvas._fontname font_size = pdf_canvas._fontsize logger.debug(f"Applied Direct track style: font={font_name}, size={font_size}") else: # OCR track or no style: Use simple font selection with element-type based styling font_name = self.font_name if self.font_registered else 'Helvetica' # Apply element-type specific styling (for OCR track) if element_type == 'title': # Titles: use larger, bold font font_size = min(font_size * 1.3, 36) # 30% larger, max 36pt pdf_canvas.setFont(font_name, font_size) logger.debug(f"Applied title style: size={font_size:.1f}") elif element_type == 'header': # Headers: slightly larger font_size = min(font_size * 1.15, 24) # 15% larger, max 24pt pdf_canvas.setFont(font_name, font_size) elif element_type == 'caption': # Captions: slightly smaller, italic if available font_size = max(font_size * 0.9, 6) # 10% smaller, min 6pt pdf_canvas.setFont(font_name, font_size) else: pdf_canvas.setFont(font_name, font_size) # Handle line breaks (split text by newlines) # OCR track: simple left-aligned rendering # Note: non_empty_lines was already calculated above for font sizing line_height = font_size * 1.2 # 120% of font size for line spacing # Draw each non-empty line (using proper line index for positioning) for i, line in enumerate(non_empty_lines): line_y = pdf_y - (i * line_height) # Calculate text width to prevent overflow text_width = pdf_canvas.stringWidth(line, font_name, font_size) # If text is too wide for bbox, scale down font for this line current_font_size = font_size if text_width > bbox_width: scale_factor = bbox_width / text_width current_font_size = font_size * scale_factor * 0.95 # 95% to add small margin current_font_size = max(current_font_size, 3) # Minimum 3pt pdf_canvas.setFont(font_name, current_font_size) # Draw text at left-aligned position (OCR track uses simple left alignment) pdf_canvas.drawString(pdf_x, line_y, line) # Reset font size for next line if text_width > bbox_width: pdf_canvas.setFont(font_name, font_size) # Debug: Draw bounding box (optional) if settings.pdf_enable_bbox_debug: pdf_canvas.setStrokeColorRGB(1, 0, 0, 0.3) # Red, semi-transparent pdf_canvas.setLineWidth(0.5) # Use already-extracted coordinates (works for all bbox formats) # Draw rectangle using the scaled coordinates pdf_x1 = ocr_x_left * scale_w pdf_y1 = page_height - ocr_y_top * scale_h pdf_x2 = ocr_x_right * scale_w pdf_y2 = page_height - ocr_y_bottom * scale_h # Draw bbox rectangle pdf_canvas.line(pdf_x1, pdf_y1, pdf_x2, pdf_y1) # top pdf_canvas.line(pdf_x2, pdf_y1, pdf_x2, pdf_y2) # right pdf_canvas.line(pdf_x2, pdf_y2, pdf_x1, pdf_y2) # bottom pdf_canvas.line(pdf_x1, pdf_y2, pdf_x1, pdf_y1) # left except Exception as e: logger.warning(f"Failed to draw text region '{text[:20]}...': {e}") def _compute_table_grid_from_cell_boxes( self, cell_boxes: List[List[float]], table_bbox: List[float], num_rows: int, num_cols: int ) -> Tuple[Optional[List[float]], Optional[List[float]]]: """ Compute column widths and row heights from cell bounding boxes. This uses the cell boxes extracted by SLANeXt to calculate the actual column widths and row heights, which provides more accurate table rendering than uniform distribution. Args: cell_boxes: List of cell bboxes [[x1,y1,x2,y2], ...] table_bbox: Table bounding box [x1,y1,x2,y2] num_rows: Number of rows in the table num_cols: Number of columns in the table Returns: Tuple of (col_widths, row_heights) or (None, None) if calculation fails """ if not cell_boxes or len(cell_boxes) < 2: return None, None try: table_x1, table_y1, table_x2, table_y2 = table_bbox table_width = table_x2 - table_x1 table_height = table_y2 - table_y1 # Collect all unique X and Y boundaries from cell boxes x_boundaries = set() y_boundaries = set() for box in cell_boxes: if len(box) >= 4: x1, y1, x2, y2 = box[:4] # Convert to relative coordinates within table x_boundaries.add(x1 - table_x1) x_boundaries.add(x2 - table_x1) y_boundaries.add(y1 - table_y1) y_boundaries.add(y2 - table_y1) # Sort boundaries x_boundaries = sorted(x_boundaries) y_boundaries = sorted(y_boundaries) # Ensure we have boundaries at table edges if x_boundaries and x_boundaries[0] > 5: x_boundaries.insert(0, 0) if x_boundaries and x_boundaries[-1] < table_width - 5: x_boundaries.append(table_width) if y_boundaries and y_boundaries[0] > 5: y_boundaries.insert(0, 0) if y_boundaries and y_boundaries[-1] < table_height - 5: y_boundaries.append(table_height) # Calculate column widths from X boundaries # Merge boundaries that are too close (< 5px) merged_x = [x_boundaries[0]] if x_boundaries else [] for x in x_boundaries[1:]: if x - merged_x[-1] > 5: merged_x.append(x) x_boundaries = merged_x # Calculate row heights from Y boundaries merged_y = [y_boundaries[0]] if y_boundaries else [] for y in y_boundaries[1:]: if y - merged_y[-1] > 5: merged_y.append(y) y_boundaries = merged_y # Calculate widths and heights col_widths = [] for i in range(len(x_boundaries) - 1): col_widths.append(x_boundaries[i + 1] - x_boundaries[i]) row_heights = [] for i in range(len(y_boundaries) - 1): row_heights.append(y_boundaries[i + 1] - y_boundaries[i]) # Validate: number of columns/rows should match expected if len(col_widths) == num_cols and len(row_heights) == num_rows: logger.info(f"[TABLE] Cell boxes grid: {num_cols} cols, {num_rows} rows") logger.debug(f"[TABLE] Col widths from cell_boxes: {[f'{w:.1f}' for w in col_widths]}") logger.debug(f"[TABLE] Row heights from cell_boxes: {[f'{h:.1f}' for h in row_heights]}") return col_widths, row_heights else: # Grid doesn't match, might be due to merged cells logger.debug( f"[TABLE] Cell boxes grid mismatch: " f"got {len(col_widths)}x{len(row_heights)}, expected {num_cols}x{num_rows}" ) # Still return the widths/heights if counts are close if abs(len(col_widths) - num_cols) <= 1 and abs(len(row_heights) - num_rows) <= 1: # Adjust to match expected count while len(col_widths) < num_cols: col_widths.append(col_widths[-1] if col_widths else table_width / num_cols) while len(col_widths) > num_cols: col_widths.pop() while len(row_heights) < num_rows: row_heights.append(row_heights[-1] if row_heights else table_height / num_rows) while len(row_heights) > num_rows: row_heights.pop() return col_widths, row_heights return None, None except Exception as e: logger.warning(f"[TABLE] Failed to compute grid from cell boxes: {e}") return None, None def draw_table_region( self, pdf_canvas: canvas.Canvas, table_element: Dict, images_metadata: List[Dict], page_height: float, scale_w: float = 1.0, scale_h: float = 1.0, result_dir: Optional[Path] = None ): """ Draw a table region by parsing HTML and rebuilding with ReportLab Table Args: pdf_canvas: ReportLab canvas object table_element: Table element dict with HTML content images_metadata: List of image metadata to find table bbox page_height: Height of page scale_w: Scale factor for X coordinates (PDF width / OCR width) scale_h: Scale factor for Y coordinates (PDF height / OCR height) result_dir: Directory containing result files (for embedded images) """ try: elem_id = table_element.get('element_id', 'unknown') use_border_only = table_element.get('_use_border_only', False) logger.info(f"[DEBUG] draw_table_region: elem_id={elem_id}, _use_border_only={use_border_only}") html_content = table_element.get('content', '') if not html_content: # Even without HTML, draw border if requested if use_border_only: self._draw_table_border_only(pdf_canvas, table_element, page_height, scale_w, scale_h) return # Apply column correction if enabled cell_boxes = table_element.get('cell_boxes', []) if (settings.table_column_correction_enabled and TABLE_COLUMN_CORRECTOR_AVAILABLE and cell_boxes): try: corrector = TableColumnCorrector( correction_threshold=settings.table_column_correction_threshold, vertical_merge_enabled=settings.vertical_fragment_merge_enabled, vertical_aspect_ratio=settings.vertical_fragment_aspect_ratio ) # Get table bbox for vertical fragment detection table_bbox = table_element.get('bbox', []) if isinstance(table_bbox, dict): table_bbox = [table_bbox['x0'], table_bbox['y0'], table_bbox['x1'], table_bbox['y1']] corrected_html, stats = corrector.correct( html=html_content, cell_boxes=cell_boxes, table_bbox=table_bbox if isinstance(table_bbox, list) and len(table_bbox) >= 4 else None ) if stats.get('column_corrections', 0) > 0: logger.info(f"[TABLE] {elem_id}: Column correction applied - {stats}") html_content = corrected_html except Exception as e: logger.warning(f"[TABLE] {elem_id}: Column correction failed: {e}, using original HTML") # Parse HTML first to get table structure for grid validation parser = HTMLTableParser() parser.feed(html_content) if not parser.tables: logger.warning("No tables found in HTML content") return # Get the first table (PP-StructureV3 usually provides one table per element) table_data = parser.tables[0] rows = table_data['rows'] if not rows: return # Calculate number of rows and columns from HTML for grid validation num_rows = len(rows) max_cols = 0 for row in rows: row_cols = sum(cell.get('colspan', 1) for cell in row['cells']) max_cols = max(max_cols, row_cols) # Check if table was rebuilt - if so, use HTML content directly was_rebuilt = table_element.get('was_rebuilt', False) cell_boxes_rendered = False # Track if we rendered borders with cell_boxes if was_rebuilt: logger.info(f"[TABLE] {elem_id}: Table was rebuilt, using HTML content directly") elif use_border_only: # Bad quality cell_boxes: skip cell_boxes rendering, use ReportLab Table with borders logger.info(f"[TABLE] {elem_id}: Bad cell_boxes quality, using ReportLab Table with borders") else: # Check if cell_boxes can produce a valid grid before rendering borders cell_boxes = table_element.get('cell_boxes', []) if cell_boxes: # Get table bbox for grid calculation temp_bbox = table_element.get('bbox', []) if isinstance(temp_bbox, dict): raw_bbox = [temp_bbox['x0'], temp_bbox['y0'], temp_bbox['x1'], temp_bbox['y1']] elif isinstance(temp_bbox, list) and len(temp_bbox) >= 4: if isinstance(temp_bbox[0], (int, float)): raw_bbox = temp_bbox[:4] else: raw_bbox = [temp_bbox[0][0], temp_bbox[0][1], temp_bbox[2][0], temp_bbox[2][1]] else: raw_bbox = None # Pre-check: can we compute a valid grid from cell_boxes? if raw_bbox: test_col_widths, test_row_heights = self._compute_table_grid_from_cell_boxes( cell_boxes, raw_bbox, num_rows, max_cols ) grid_valid = test_col_widths is not None and test_row_heights is not None if grid_valid: logger.info(f"[TABLE] Grid validation passed, rendering borders with cell_boxes") success = self._draw_table_with_cell_boxes( pdf_canvas, table_element, page_height, scale_w, scale_h, result_dir ) if success: cell_boxes_rendered = True logger.info("[TABLE] cell_boxes rendered borders, continuing with text-only ReportLab Table") else: logger.info("[TABLE] cell_boxes rendering failed, using ReportLab Table with borders") else: # Grid mismatch: try cellboxes-first rendering if enabled if settings.table_rendering_prefer_cellboxes: logger.info(f"[TABLE] Grid mismatch, trying cellboxes-first rendering") from app.services.pdf_table_renderer import TableRenderer, TableRenderConfig renderer = TableRenderer(TableRenderConfig()) success = renderer.render_from_cellboxes_grid( pdf_canvas, cell_boxes, html_content, tuple(raw_bbox), page_height, scale_w, scale_h, row_threshold=settings.table_cellboxes_row_threshold, col_threshold=settings.table_cellboxes_col_threshold ) if success: logger.info("[TABLE] cellboxes-first rendering succeeded, skipping HTML-based rendering") return # Table fully rendered, exit early else: logger.info("[TABLE] cellboxes-first rendering failed, falling back to HTML-based") else: logger.info(f"[TABLE] Grid validation failed (mismatch), using ReportLab Table with borders") else: logger.info("[TABLE] No valid bbox for grid validation, using ReportLab Table with borders") # Get bbox directly from table element table_bbox = table_element.get('bbox') # If no bbox directly, check for bbox_polygon if not table_bbox: bbox_polygon = table_element.get('bbox_polygon') if bbox_polygon and len(bbox_polygon) >= 4: # Convert polygon format to simple bbox [x0, y0, x1, y1] table_bbox = [ bbox_polygon[0][0], # x0 bbox_polygon[0][1], # y0 bbox_polygon[2][0], # x1 bbox_polygon[2][1] # y1 ] if not table_bbox: logger.warning(f"No bbox found for table element") return # Handle different bbox formats if isinstance(table_bbox, dict): # Dict format from UnifiedDocument: {"x0": ..., "y0": ..., "x1": ..., "y1": ...} if 'x0' in table_bbox and 'y0' in table_bbox and 'x1' in table_bbox and 'y1' in table_bbox: ocr_x_left_raw = float(table_bbox['x0']) ocr_y_top_raw = float(table_bbox['y0']) ocr_x_right_raw = float(table_bbox['x1']) ocr_y_bottom_raw = float(table_bbox['y1']) else: logger.error(f"Dict bbox missing required keys (x0, y0, x1, y1): {table_bbox}") return elif isinstance(table_bbox, list) and len(table_bbox) == 4: # Simple bbox format [x0, y0, x1, y1] if isinstance(table_bbox[0], (int, float)): ocr_x_left_raw = table_bbox[0] ocr_y_top_raw = table_bbox[1] ocr_x_right_raw = table_bbox[2] ocr_y_bottom_raw = table_bbox[3] # Polygon format [[x,y], [x,y], [x,y], [x,y]] elif isinstance(table_bbox[0], list): ocr_x_left_raw = table_bbox[0][0] ocr_y_top_raw = table_bbox[0][1] ocr_x_right_raw = table_bbox[2][0] ocr_y_bottom_raw = table_bbox[2][1] else: logger.error(f"Unexpected bbox format: {table_bbox}") return else: logger.error(f"Invalid table_bbox format: {table_bbox}") return logger.info(f"[表格] OCR原始座標: L={ocr_x_left_raw:.0f}, T={ocr_y_top_raw:.0f}, R={ocr_x_right_raw:.0f}, B={ocr_y_bottom_raw:.0f}") # Apply scaling ocr_x_left = ocr_x_left_raw * scale_w ocr_y_top = ocr_y_top_raw * scale_h ocr_x_right = ocr_x_right_raw * scale_w ocr_y_bottom = ocr_y_bottom_raw * scale_h table_width = abs(ocr_x_right - ocr_x_left) table_height = abs(ocr_y_bottom - ocr_y_top) # Transform coordinates pdf_x = ocr_x_left pdf_y = page_height - ocr_y_bottom # Build table data for ReportLab with proper colspan/rowspan handling # num_rows and max_cols already calculated above for grid validation logger.info(f"[表格] {num_rows}行x{max_cols}列 → PDF位置: ({pdf_x:.1f}, {pdf_y:.1f}), 寬x高: {table_width:.0f}x{table_height:.0f}") # Create a grid to track occupied cells (for rowspan handling) # occupied[row][col] = True if cell is occupied by a span from above occupied = [[False] * max_cols for _ in range(num_rows)] # Build the 2D data array and collect span commands reportlab_data = [] span_commands = [] for row_idx, row in enumerate(rows): row_data = [''] * max_cols col_idx = 0 for cell in row['cells']: # Skip occupied cells (from rowspan above) while col_idx < max_cols and occupied[row_idx][col_idx]: col_idx += 1 if col_idx >= max_cols: break text = cell['text'].strip() colspan = cell.get('colspan', 1) rowspan = cell.get('rowspan', 1) # Place text in the top-left cell of the span row_data[col_idx] = text # Mark cells as occupied for rowspan for r in range(row_idx, min(row_idx + rowspan, num_rows)): for c in range(col_idx, min(col_idx + colspan, max_cols)): if r > row_idx or c > col_idx: occupied[r][c] = True # Add SPAN command if cell spans multiple rows/cols if colspan > 1 or rowspan > 1: span_end_col = min(col_idx + colspan - 1, max_cols - 1) span_end_row = min(row_idx + rowspan - 1, num_rows - 1) span_commands.append(('SPAN', (col_idx, row_idx), (span_end_col, span_end_row))) col_idx += colspan reportlab_data.append(row_data) # Calculate column widths and row heights # First, try to use cell_boxes if available for more accurate layout cell_boxes = table_element.get('cell_boxes') raw_table_bbox = [ocr_x_left_raw, ocr_y_top_raw, ocr_x_right_raw, ocr_y_bottom_raw] computed_col_widths = None computed_row_heights = None if cell_boxes: cell_boxes_source = table_element.get('cell_boxes_source', 'unknown') logger.info(f"[TABLE] Using {len(cell_boxes)} cell boxes from {cell_boxes_source}") computed_col_widths, computed_row_heights = self._compute_table_grid_from_cell_boxes( cell_boxes, raw_table_bbox, num_rows, max_cols ) # Use computed widths if available, otherwise fall back to equal distribution if computed_col_widths: # Scale col_widths to PDF coordinates col_widths = [w * scale_w for w in computed_col_widths] logger.info(f"[TABLE] Using cell_boxes col widths (scaled)") else: col_widths = [table_width / max_cols] * max_cols logger.info(f"[TABLE] Using equal distribution col widths: {table_width/max_cols:.1f} each") # Row heights - ALWAYS use to ensure table fits bbox properly # Use computed heights from cell_boxes, or uniform distribution as fallback if computed_row_heights: # Scale row_heights to PDF coordinates row_heights = [h * scale_h for h in computed_row_heights] logger.info(f"[TABLE] Using cell_boxes row heights (scaled)") else: # Uniform distribution based on table bbox - ensures table fills its allocated space row_heights = [table_height / num_rows] * num_rows logger.info(f"[TABLE] Using uniform row heights: {table_height/num_rows:.1f} each") # Create ReportLab Table # Use smaller font to fit content with auto-wrap font_size = 8 # Fixed reasonable font size for table content # Create paragraph style for text wrapping in cells cell_style = ParagraphStyle( 'CellStyle', fontName=self.font_name if self.font_registered else 'Helvetica', fontSize=font_size, leading=font_size * 1.2, alignment=TA_CENTER, wordWrap='CJK', # Better wrapping for Chinese text ) # Convert text to Paragraph objects for auto-wrapping for row_idx, row_data in enumerate(reportlab_data): for col_idx, cell_text in enumerate(row_data): if cell_text: # Escape HTML special characters and create Paragraph escaped_text = cell_text.replace('&', '&').replace('<', '<').replace('>', '>') reportlab_data[row_idx][col_idx] = Paragraph(escaped_text, cell_style) # Create table with col widths and row heights # Always use row_heights to ensure table fits bbox properly table = Table(reportlab_data, colWidths=col_widths, rowHeights=row_heights) logger.info(f"[TABLE] Created with {len(col_widths)} cols, {len(row_heights)} rows") # Apply table style # If cell_boxes rendered borders, skip GRID style (text-only rendering) style_commands = [ ('FONT', (0, 0), (-1, -1), self.font_name if self.font_registered else 'Helvetica', font_size), ('VALIGN', (0, 0), (-1, -1), 'MIDDLE'), ('ALIGN', (0, 0), (-1, -1), 'CENTER'), ('LEFTPADDING', (0, 0), (-1, -1), 2), ('RIGHTPADDING', (0, 0), (-1, -1), 2), ('TOPPADDING', (0, 0), (-1, -1), 2), ('BOTTOMPADDING', (0, 0), (-1, -1), 2), ] # Only add GRID if cell_boxes didn't render borders if not cell_boxes_rendered: style_commands.insert(1, ('GRID', (0, 0), (-1, -1), 0.5, colors.black)) logger.info("[TABLE] Adding GRID style (cell_boxes not used)") else: logger.info("[TABLE] Skipping GRID style (cell_boxes rendered borders)") style = TableStyle(style_commands) # Add header style if first row has headers if rows and rows[0]['cells'] and rows[0]['cells'][0].get('is_header'): style.add('BACKGROUND', (0, 0), (-1, 0), colors.lightgrey) style.add('FONT', (0, 0), (-1, 0), self.font_name if self.font_registered else 'Helvetica-Bold', font_size) # Add span commands for merged cells for span_cmd in span_commands: style.add(*span_cmd) table.setStyle(style) logger.info(f"[表格] 套用 {len(span_commands)} 個合併儲存格 (SPAN)") # Calculate actual table size after wrapping actual_width, actual_height = table.wrapOn(pdf_canvas, table_width, table_height) logger.info(f"[表格] 目標尺寸: {table_width:.0f}x{table_height:.0f}, 實際尺寸: {actual_width:.0f}x{actual_height:.0f}") # Scale table to fit bbox if it exceeds the target size scale_x = table_width / actual_width if actual_width > table_width else 1.0 scale_y = table_height / actual_height if actual_height > table_height else 1.0 scale_factor = min(scale_x, scale_y) # Use smaller scale to fit both dimensions # Calculate the table top position in PDF coordinates # ReportLab uses bottom-left origin, so we need to position from TOP pdf_y_top = page_height - ocr_y_top # Top of table in PDF coords # Calculate the actual bottom position based on scaled height # Table should be positioned so its TOP aligns with the bbox top scaled_height = actual_height * scale_factor pdf_y_bottom = pdf_y_top - scaled_height # Bottom of scaled table logger.info(f"[表格] PDF座標: top={pdf_y_top:.0f}, bottom={pdf_y_bottom:.0f}, scaled_height={scaled_height:.0f}") if scale_factor < 1.0: logger.info(f"[表格] 縮放比例: {scale_factor:.2f} (需要縮小以適應 bbox)") # Apply scaling transformation pdf_canvas.saveState() pdf_canvas.translate(pdf_x, pdf_y_bottom) pdf_canvas.scale(scale_factor, scale_factor) # Draw at origin since we've already translated table.drawOn(pdf_canvas, 0, 0) pdf_canvas.restoreState() else: # Draw table at position without scaling # pdf_y should be the bottom of the table table.drawOn(pdf_canvas, pdf_x, pdf_y_bottom) logger.info(f"Drew table at ({pdf_x:.0f}, {pdf_y_bottom:.0f}) size {table_width:.0f}x{scaled_height:.0f} with {len(rows)} rows") # Draw embedded images (images detected inside the table region) embedded_images = table_element.get('embedded_images', []) if embedded_images and result_dir: logger.info(f"[TABLE] Drawing {len(embedded_images)} embedded images") for emb_img in embedded_images: self._draw_embedded_image( pdf_canvas, emb_img, page_height, result_dir, scale_w, scale_h ) except Exception as e: logger.warning(f"Failed to draw table region: {e}") import traceback traceback.print_exc() def _draw_embedded_image( self, pdf_canvas: canvas.Canvas, emb_img: Dict, page_height: float, result_dir: Path, scale_w: float = 1.0, scale_h: float = 1.0 ): """Draw an embedded image inside a table region.""" try: # Get image path saved_path = emb_img.get('saved_path', '') if not saved_path: return # Construct full path image_path = result_dir / saved_path if not image_path.exists(): image_path = result_dir / Path(saved_path).name if not image_path.exists(): logger.warning(f"Embedded image not found: {saved_path}") return # Get bbox from embedded image data bbox = emb_img.get('bbox', []) if not bbox or len(bbox) < 4: logger.warning(f"No bbox for embedded image: {saved_path}") return # Calculate position (bbox is [x0, y0, x1, y1]) x0, y0, x1, y1 = bbox[0], bbox[1], bbox[2], bbox[3] # Apply scaling x0_scaled = x0 * scale_w y0_scaled = y0 * scale_h x1_scaled = x1 * scale_w y1_scaled = y1 * scale_h width = x1_scaled - x0_scaled height = y1_scaled - y0_scaled # Transform Y coordinate (ReportLab uses bottom-left origin) pdf_x = x0_scaled pdf_y = page_height - y1_scaled # Draw the image from reportlab.lib.utils import ImageReader img_reader = ImageReader(str(image_path)) pdf_canvas.drawImage( img_reader, pdf_x, pdf_y, width, height, preserveAspectRatio=True, mask='auto' ) logger.info(f"Drew embedded image at ({pdf_x:.0f}, {pdf_y:.0f}) size {width:.0f}x{height:.0f}") except Exception as e: logger.warning(f"Failed to draw embedded image: {e}") def _normalize_cell_boxes_to_grid( self, cell_boxes: List[List[float]], threshold: float = 10.0 ) -> List[List[float]]: """ Normalize cell boxes to create a proper aligned grid. Groups nearby coordinates and snaps them to a common value, eliminating the 2-11 pixel variations that cause skewed tables. Args: cell_boxes: List of cell bboxes [[x1,y1,x2,y2], ...] threshold: Maximum distance to consider coordinates as "same line" Returns: Normalized cell_boxes with aligned coordinates """ if not cell_boxes or len(cell_boxes) < 2: return cell_boxes # Collect all X and Y coordinates x_coords = [] # (value, box_idx, is_x1) y_coords = [] # (value, box_idx, is_y1) for i, box in enumerate(cell_boxes): x1, y1, x2, y2 = box[0], box[1], box[2], box[3] x_coords.append((x1, i, True)) # x1 (left) x_coords.append((x2, i, False)) # x2 (right) y_coords.append((y1, i, True)) # y1 (top) y_coords.append((y2, i, False)) # y2 (bottom) def cluster_and_normalize(coords, threshold): """Cluster nearby coordinates and return mapping to normalized values.""" if not coords: return {} # Sort by value sorted_coords = sorted(coords, key=lambda x: x[0]) # Cluster nearby values clusters = [] current_cluster = [sorted_coords[0]] for coord in sorted_coords[1:]: if coord[0] - current_cluster[-1][0] <= threshold: current_cluster.append(coord) else: clusters.append(current_cluster) current_cluster = [coord] clusters.append(current_cluster) # Create mapping: (box_idx, is_first) -> normalized value mapping = {} for cluster in clusters: # Use average of cluster as normalized value avg_value = sum(c[0] for c in cluster) / len(cluster) for _, box_idx, is_first in cluster: mapping[(box_idx, is_first)] = avg_value return mapping x_mapping = cluster_and_normalize(x_coords, threshold) y_mapping = cluster_and_normalize(y_coords, threshold) # Create normalized cell boxes normalized_boxes = [] for i, box in enumerate(cell_boxes): x1_norm = x_mapping.get((i, True), box[0]) x2_norm = x_mapping.get((i, False), box[2]) y1_norm = y_mapping.get((i, True), box[1]) y2_norm = y_mapping.get((i, False), box[3]) normalized_boxes.append([x1_norm, y1_norm, x2_norm, y2_norm]) logger.debug(f"[TABLE] Normalized {len(cell_boxes)} cell boxes to grid") return normalized_boxes def _draw_table_border_only( self, pdf_canvas: canvas.Canvas, table_element: Dict, page_height: float, scale_w: float = 1.0, scale_h: float = 1.0 ): """ Draw only the outer border of a table (for tables with bad cell_boxes quality). Text inside the table will be rendered using raw OCR positions. Args: pdf_canvas: ReportLab canvas object table_element: Table element dict page_height: Height of page in PDF coordinates scale_w: Scale factor for X coordinates scale_h: Scale factor for Y coordinates """ table_bbox = table_element.get('bbox', []) if not table_bbox or len(table_bbox) < 4: return element_id = table_element.get('element_id', 'unknown') # Handle different bbox formats if isinstance(table_bbox, dict): x0, y0, x1, y1 = table_bbox['x0'], table_bbox['y0'], table_bbox['x1'], table_bbox['y1'] elif isinstance(table_bbox[0], (int, float)): x0, y0, x1, y1 = table_bbox[0], table_bbox[1], table_bbox[2], table_bbox[3] else: return # Apply scaling pdf_x0 = x0 * scale_w pdf_y0 = y0 * scale_h pdf_x1 = x1 * scale_w pdf_y1 = y1 * scale_h # Convert to PDF coordinates (flip Y) pdf_top = page_height - pdf_y0 pdf_bottom = page_height - pdf_y1 width = pdf_x1 - pdf_x0 height = pdf_y1 - pdf_y0 # Draw outer border only pdf_canvas.setStrokeColor(colors.black) pdf_canvas.setLineWidth(0.5) pdf_canvas.rect(pdf_x0, pdf_bottom, width, height, stroke=1, fill=0) logger.info(f"[TABLE] {element_id}: Drew border only (bad cell_boxes quality)") def _check_cell_boxes_quality(self, cell_boxes: List, element_id: str = "") -> str: """ Check the quality of cell_boxes to determine rendering strategy. Args: cell_boxes: List of cell bounding boxes element_id: Optional element ID for logging Returns: 'good' if cell_boxes form a proper grid, 'bad' otherwise """ # If quality check is disabled, always return 'good' to use pure PP-Structure output if not settings.table_quality_check_enabled: logger.debug(f"[TABLE QUALITY] {element_id}: good - quality check disabled (pure PP-Structure mode)") return 'good' if not cell_boxes or len(cell_boxes) < 2: logger.debug(f"[TABLE QUALITY] {element_id}: bad - too few cells ({len(cell_boxes) if cell_boxes else 0})") return 'bad' # No cell_boxes or too few # Count overlapping cell pairs overlap_count = 0 for i, box1 in enumerate(cell_boxes): for j, box2 in enumerate(cell_boxes): if i >= j: continue if not isinstance(box1, (list, tuple)) or len(box1) < 4: continue if not isinstance(box2, (list, tuple)) or len(box2) < 4: continue x_overlap = box1[0] < box2[2] and box1[2] > box2[0] y_overlap = box1[1] < box2[3] and box1[3] > box2[1] if x_overlap and y_overlap: overlap_count += 1 total_pairs = len(cell_boxes) * (len(cell_boxes) - 1) // 2 overlap_ratio = overlap_count / total_pairs if total_pairs > 0 else 0 # Relaxed threshold: 20% overlap instead of 10% to allow more tables through # This is because PP-StructureV3's cell detection sometimes has slight overlaps if overlap_ratio > 0.20: logger.info(f"[TABLE QUALITY] {element_id}: bad - overlap ratio {overlap_ratio:.2%} > 20%") return 'bad' logger.debug(f"[TABLE QUALITY] {element_id}: good - {len(cell_boxes)} cells, overlap {overlap_ratio:.2%}") return 'good' def _draw_table_with_cell_boxes( self, pdf_canvas: canvas.Canvas, table_element: Dict, page_height: float, scale_w: float = 1.0, scale_h: float = 1.0, result_dir: Optional[Path] = None ): """ Draw table borders using cell_boxes for accurate positioning. LAYERED RENDERING APPROACH: - This method ONLY draws cell borders and embedded images - Text is rendered separately using raw OCR positions (via GapFillingService) - This decouples visual structure (borders) from content (text) FALLBACK: If cell_boxes are incomplete, always draws the outer table border using the table's bbox to ensure table boundaries are visible. Args: pdf_canvas: ReportLab canvas object table_element: Table element dict with cell_boxes page_height: Height of page in PDF coordinates scale_w: Scale factor for X coordinates scale_h: Scale factor for Y coordinates result_dir: Directory containing result files (for embedded images) """ try: cell_boxes = table_element.get('cell_boxes', []) table_bbox = table_element.get('bbox', []) # Check cell_boxes quality - skip if they don't form a proper grid if cell_boxes and len(cell_boxes) > 2: # Count overlapping cell pairs overlap_count = 0 for i, box1 in enumerate(cell_boxes): for j, box2 in enumerate(cell_boxes): if i >= j: continue x_overlap = box1[0] < box2[2] and box1[2] > box2[0] y_overlap = box1[1] < box2[3] and box1[3] > box2[1] if x_overlap and y_overlap: overlap_count += 1 # If more than 25% of cell pairs overlap, cell_boxes are unreliable # Increased from 10% to 25% to allow more tables to use cell_boxes rendering # which provides better visual fidelity than ReportLab Table fallback total_pairs = len(cell_boxes) * (len(cell_boxes) - 1) // 2 overlap_ratio = overlap_count / total_pairs if total_pairs > 0 else 0 if overlap_ratio > 0.25: logger.warning( f"[TABLE] Skipping cell_boxes rendering: {overlap_count}/{total_pairs} " f"({overlap_ratio:.1%}) cell pairs overlap - using ReportLab Table fallback" ) return False # Return False to trigger ReportLab Table fallback if not cell_boxes: # Fallback: draw outer border only when no cell_boxes if table_bbox and len(table_bbox) >= 4: # Handle different bbox formats (list or dict) if isinstance(table_bbox, dict): tx1 = float(table_bbox.get('x0', 0)) ty1 = float(table_bbox.get('y0', 0)) tx2 = float(table_bbox.get('x1', 0)) ty2 = float(table_bbox.get('y1', 0)) else: tx1, ty1, tx2, ty2 = table_bbox[:4] # Apply scaling tx1_scaled = tx1 * scale_w ty1_scaled = ty1 * scale_h tx2_scaled = tx2 * scale_w ty2_scaled = ty2 * scale_h table_width = tx2_scaled - tx1_scaled table_height = ty2_scaled - ty1_scaled # Transform Y coordinate (PDF uses bottom-left origin) pdf_x = tx1_scaled pdf_y = page_height - ty2_scaled # Bottom of table in PDF coords # Draw outer table border (slightly thicker for visibility) pdf_canvas.setStrokeColor(colors.black) pdf_canvas.setLineWidth(1.0) pdf_canvas.rect(pdf_x, pdf_y, table_width, table_height, stroke=1, fill=0) logger.info(f"[TABLE] Drew outer table border at [{int(tx1)},{int(ty1)},{int(tx2)},{int(ty2)}]") logger.warning("[TABLE] No cell_boxes available, only outer border drawn") # Still draw embedded images even without cell borders embedded_images = table_element.get('embedded_images', []) if embedded_images and result_dir: for emb_img in embedded_images: self._draw_embedded_image( pdf_canvas, emb_img, page_height, result_dir, scale_w, scale_h ) return True # Outer border drawn successfully # Normalize cell boxes to create aligned grid cell_boxes = self._normalize_cell_boxes_to_grid(cell_boxes) logger.info(f"[TABLE] Drawing {len(cell_boxes)} cells using grid lines (avoiding duplicates)") # Collect unique grid lines to avoid drawing duplicate/overlapping lines h_lines = set() # Horizontal lines: (y, x_start, x_end) v_lines = set() # Vertical lines: (x, y_start, y_end) for box in cell_boxes: x1, y1, x2, y2 = box[0], box[1], box[2], box[3] # Apply scaling x1_s = x1 * scale_w y1_s = y1 * scale_h x2_s = x2 * scale_w y2_s = y2 * scale_h # Round to 1 decimal place to help with deduplication x1_s, y1_s, x2_s, y2_s = round(x1_s, 1), round(y1_s, 1), round(x2_s, 1), round(y2_s, 1) # Add horizontal lines (top and bottom of cell) h_lines.add((y1_s, x1_s, x2_s)) # Top line h_lines.add((y2_s, x1_s, x2_s)) # Bottom line # Add vertical lines (left and right of cell) v_lines.add((x1_s, y1_s, y2_s)) # Left line v_lines.add((x2_s, y1_s, y2_s)) # Right line # Draw unique horizontal lines pdf_canvas.setStrokeColor(colors.black) pdf_canvas.setLineWidth(0.5) for y, x_start, x_end in h_lines: pdf_y = page_height - y # Transform Y coordinate pdf_canvas.line(x_start, pdf_y, x_end, pdf_y) # Draw unique vertical lines for x, y_start, y_end in v_lines: pdf_y_start = page_height - y_start pdf_y_end = page_height - y_end pdf_canvas.line(x, pdf_y_start, x, pdf_y_end) logger.info(f"[TABLE] Drew {len(h_lines)} horizontal + {len(v_lines)} vertical grid lines") # Draw embedded images embedded_images = table_element.get('embedded_images', []) if embedded_images and result_dir: logger.info(f"[TABLE] Drawing {len(embedded_images)} embedded images") for emb_img in embedded_images: self._draw_embedded_image( pdf_canvas, emb_img, page_height, result_dir, scale_w, scale_h ) return True except Exception as e: logger.warning(f"[TABLE] Failed to draw cell borders: {e}") import traceback traceback.print_exc() return False def draw_image_region( self, pdf_canvas: canvas.Canvas, region: Dict, page_height: float, result_dir: Path, scale_w: float = 1.0, scale_h: float = 1.0 ): """ Draw an image region by embedding the extracted image Handles images extracted by PP-StructureV3 (tables, figures, charts, etc.) Args: pdf_canvas: ReportLab canvas object region: Image metadata dict with image_path and bbox page_height: Height of page (for coordinate transformation) result_dir: Directory containing result files scale_w: Scale factor for X coordinates (PDF width / OCR width) scale_h: Scale factor for Y coordinates (PDF height / OCR height) """ try: image_path_str = region.get('image_path', '') if not image_path_str: return # Construct full path to image # saved_path is relative to result_dir (e.g., "imgs/element_id.png") image_path = result_dir / image_path_str # Fallback for legacy data if not image_path.exists(): image_path = result_dir / Path(image_path_str).name if not image_path.exists(): logger.warning(f"Image not found: {image_path_str} (in {result_dir})") return # Get bbox for positioning bbox = region.get('bbox', []) if not bbox: logger.warning(f"No bbox for image {image_path_str}") return # Handle different bbox formats if isinstance(bbox, dict): # Dict format from UnifiedDocument: {"x0": ..., "y0": ..., "x1": ..., "y1": ...} if 'x0' in bbox and 'y0' in bbox and 'x1' in bbox and 'y1' in bbox: ocr_x_left_raw = float(bbox['x0']) ocr_y_top_raw = float(bbox['y0']) ocr_x_right_raw = float(bbox['x1']) ocr_y_bottom_raw = float(bbox['y1']) else: logger.warning(f"Dict bbox missing required keys for image: {bbox}") return elif isinstance(bbox, list): if len(bbox) < 4: logger.warning(f"List bbox too short for image: {bbox}") return # Polygon format [[x,y], [x,y], [x,y], [x,y]] if isinstance(bbox[0], list): ocr_x_left_raw = bbox[0][0] ocr_y_top_raw = bbox[0][1] ocr_x_right_raw = bbox[2][0] ocr_y_bottom_raw = bbox[2][1] # Simple list format [x0, y0, x1, y1] elif isinstance(bbox[0], (int, float)): ocr_x_left_raw = bbox[0] ocr_y_top_raw = bbox[1] ocr_x_right_raw = bbox[2] ocr_y_bottom_raw = bbox[3] else: logger.warning(f"Unexpected bbox list format for image: {bbox}") return else: logger.warning(f"Invalid bbox format for image: {bbox}") return logger.info(f"[圖片] '{image_path_str}' OCR原始座標: L={ocr_x_left_raw:.0f}, T={ocr_y_top_raw:.0f}, R={ocr_x_right_raw:.0f}, B={ocr_y_bottom_raw:.0f}") # Apply scaling ocr_x_left = ocr_x_left_raw * scale_w ocr_y_top = ocr_y_top_raw * scale_h ocr_x_right = ocr_x_right_raw * scale_w ocr_y_bottom = ocr_y_bottom_raw * scale_h # Calculate bbox dimensions (after scaling) bbox_width = abs(ocr_x_right - ocr_x_left) bbox_height = abs(ocr_y_bottom - ocr_y_top) # Transform coordinates: OCR (top-left origin) → PDF (bottom-left origin) # CRITICAL: Y-axis flip! # For images, we position at bottom-left corner pdf_x_left = ocr_x_left pdf_y_bottom = page_height - ocr_y_bottom # Flip Y-axis logger.info(f"[圖片] '{image_path_str}' → PDF位置: ({pdf_x_left:.1f}, {pdf_y_bottom:.1f}), 寬x高: {bbox_width:.0f}x{bbox_height:.0f}") # Draw image using ReportLab # drawImage expects: (path, x, y, width, height) # where (x, y) is the bottom-left corner of the image pdf_canvas.drawImage( str(image_path), pdf_x_left, pdf_y_bottom, width=bbox_width, height=bbox_height, preserveAspectRatio=True, mask='auto' # Handle transparency ) logger.info(f"[圖片] ✓ 成功繪製 '{image_path_str}'") except Exception as e: logger.warning(f"Failed to draw image region: {e}") def generate_layout_pdf( self, json_path: Path, output_path: Path, source_file_path: Optional[Path] = None ) -> bool: """ Generate layout-preserving PDF from OCR JSON data Args: json_path: Path to OCR JSON file output_path: Path to save generated PDF source_file_path: Optional path to original source file for dimension extraction Returns: True if successful, False otherwise """ try: # Load JSON data ocr_data = self.load_ocr_json(json_path) if not ocr_data: return False # Check if this is new UnifiedDocument format (has 'pages' with elements) # vs old OCR format (has 'text_regions') if 'pages' in ocr_data and isinstance(ocr_data.get('pages'), list): # New UnifiedDocument format - convert and use Direct track rendering logger.info("Detected UnifiedDocument JSON format, using Direct track rendering") unified_doc = self._json_to_unified_document(ocr_data, json_path.parent) if unified_doc: return self.generate_from_unified_document( unified_doc=unified_doc, output_path=output_path, source_file_path=source_file_path ) else: logger.error("Failed to convert JSON to UnifiedDocument") return False else: # Old OCR format - use legacy generation logger.info("Detected legacy OCR JSON format, using OCR track rendering") return self._generate_pdf_from_data( ocr_data=ocr_data, output_path=output_path, source_file_path=source_file_path, json_parent_dir=json_path.parent ) except Exception as e: logger.error(f"Failed to generate PDF: {e}") import traceback traceback.print_exc() return False def _json_to_unified_document(self, json_data: Dict, result_dir: Path) -> Optional['UnifiedDocument']: """ Convert JSON dict to UnifiedDocument object. Args: json_data: Loaded JSON dictionary in UnifiedDocument format result_dir: Directory containing image files Returns: UnifiedDocument object or None if conversion fails """ try: from datetime import datetime # Parse metadata metadata_dict = json_data.get('metadata', {}) # Parse processing track track_str = metadata_dict.get('processing_track', 'direct') try: processing_track = ProcessingTrack(track_str) except ValueError: processing_track = ProcessingTrack.DIRECT # Create DocumentMetadata metadata = DocumentMetadata( filename=metadata_dict.get('filename', ''), file_type=metadata_dict.get('file_type', 'pdf'), file_size=metadata_dict.get('file_size', 0), created_at=datetime.fromisoformat(metadata_dict.get('created_at', datetime.now().isoformat()).replace('Z', '+00:00')), processing_track=processing_track, processing_time=metadata_dict.get('processing_time', 0), language=metadata_dict.get('language'), title=metadata_dict.get('title'), author=metadata_dict.get('author'), subject=metadata_dict.get('subject'), keywords=metadata_dict.get('keywords'), producer=metadata_dict.get('producer'), creator=metadata_dict.get('creator'), creation_date=datetime.fromisoformat(metadata_dict['creation_date'].replace('Z', '+00:00')) if metadata_dict.get('creation_date') else None, modification_date=datetime.fromisoformat(metadata_dict['modification_date'].replace('Z', '+00:00')) if metadata_dict.get('modification_date') else None, ) # Parse pages pages = [] for page_dict in json_data.get('pages', []): # Parse page dimensions dims = page_dict.get('dimensions', {}) if not dims: # Fallback dimensions dims = {'width': 595.32, 'height': 841.92} dimensions = Dimensions( width=dims.get('width', 595.32), height=dims.get('height', 841.92), dpi=dims.get('dpi') ) # Parse elements elements = [] for elem_dict in page_dict.get('elements', []): element = self._json_to_document_element(elem_dict) if element: elements.append(element) page = Page( page_number=page_dict.get('page_number', 1), dimensions=dimensions, elements=elements, metadata=page_dict.get('metadata', {}) ) pages.append(page) # Create UnifiedDocument unified_doc = UnifiedDocument( document_id=json_data.get('document_id', ''), metadata=metadata, pages=pages, processing_errors=json_data.get('processing_errors', []) ) logger.info(f"Converted JSON to UnifiedDocument: {len(pages)} pages, track={processing_track.value}") return unified_doc except Exception as e: logger.error(f"Failed to convert JSON to UnifiedDocument: {e}") import traceback traceback.print_exc() return None def _json_to_document_element(self, elem_dict: Dict) -> Optional['DocumentElement']: """ Convert JSON dict to DocumentElement. Args: elem_dict: Element dictionary from JSON Returns: DocumentElement or None if conversion fails """ try: # Parse element type type_str = elem_dict.get('type', 'text') try: elem_type = ElementType(type_str) except ValueError: # Fallback to TEXT for unknown types elem_type = ElementType.TEXT logger.warning(f"Unknown element type '{type_str}', falling back to TEXT") # Content-based HTML table detection: reclassify text elements with HTML table content content = elem_dict.get('content', '') if elem_type == ElementType.TEXT and isinstance(content, str) and ' bool: """ Fallback detection for list items not marked with ElementType.LIST_ITEM. Checks metadata and text patterns to identify list items. Args: element: Document element to check Returns: True if element appears to be a list item """ # Skip if already categorized as table or image if element.type in [ElementType.TABLE, ElementType.IMAGE, ElementType.FIGURE, ElementType.CHART, ElementType.DIAGRAM]: return False # Check metadata for list-related fields if element.metadata: # Check for list_level metadata if 'list_level' in element.metadata: return True # Check for parent_item (indicates list hierarchy) if 'parent_item' in element.metadata: return True # Check for children (could be parent list item) if 'children' in element.metadata and element.metadata['children']: return True # Check text content for list patterns if element.is_text: text = element.get_text().lstrip() # Ordered list pattern: starts with number followed by . or ) if re.match(r'^\d+[\.\)]\s', text): return True # Unordered list pattern: starts with bullet character if re.match(r'^[•·▪▫◦‣⁃\-\*]\s', text): return True return False def _draw_list_elements_direct( self, pdf_canvas: canvas.Canvas, list_elements: List['DocumentElement'], page_height: float ): """ Draw list elements with proper sequential numbering and formatting. This method processes all list items on a page, groups them into lists, and assigns proper sequential numbering to ordered lists. Args: pdf_canvas: ReportLab canvas object list_elements: List of LIST_ITEM elements page_height: Page height for coordinate transformation """ if not list_elements: return # Sort list items by position (top to bottom, left to right) sorted_items = sorted(list_elements, key=lambda e: (e.bbox.y0, e.bbox.x0)) # Group list items into lists based on proximity and level list_groups = [] current_group = [] prev_y = None prev_level = None max_gap = 30 # Maximum vertical gap between items in same list (in points) for item in sorted_items: level = item.metadata.get('list_level', 0) if item.metadata else 0 y_pos = item.bbox.y0 # Check if this item belongs to current group if current_group and prev_y is not None: gap = abs(y_pos - prev_y) # Start new group if gap is too large or level changed significantly if gap > max_gap or (prev_level is not None and level != prev_level): list_groups.append(current_group) current_group = [] current_group.append(item) prev_y = y_pos prev_level = level if current_group: list_groups.append(current_group) # Process each list group for group in list_groups: # Detect list type from first item first_item = group[0] text_content = first_item.get_text() text_stripped = text_content.lstrip() list_type = None list_counter = 1 # Determine list type if re.match(r'^\d+[\.\)]\s', text_stripped): list_type = 'ordered' # Extract starting number match = re.match(r'^(\d+)[\.\)]\s', text_stripped) if match: list_counter = int(match.group(1)) elif re.match(r'^[•·▪▫◦‣⁃]\s', text_stripped): list_type = 'unordered' # Draw each item in the group with proper spacing # Track cumulative Y offset to apply spacing_after between items cumulative_y_offset = 0 for item_idx, item in enumerate(group): # Prepare list marker based on type if list_type == 'ordered': list_marker = f"{list_counter}. " list_counter += 1 elif list_type == 'unordered': list_marker = "• " else: list_marker = "" # No marker if type unknown # Store list marker in item metadata for _draw_text_element_direct if not item.metadata: item.metadata = {} item.metadata['_list_marker'] = list_marker item.metadata['_list_type'] = list_type # Add default list item spacing if not specified # This ensures consistent spacing between list items desired_spacing_after = item.metadata.get('spacing_after', 0) if desired_spacing_after == 0: # Default list item spacing: 3 points between items (except last item) if item_idx < len(group) - 1: desired_spacing_after = 3.0 item.metadata['spacing_after'] = desired_spacing_after # Draw the list item with cumulative Y offset self._draw_text_element_direct(pdf_canvas, item, page_height, y_offset=cumulative_y_offset) # Calculate spacing to add after this item if item_idx < len(group) - 1 and desired_spacing_after > 0: next_item = group[item_idx + 1] # Calculate actual vertical gap between items (in document coordinates) # Note: Y increases downward in document coordinates actual_gap = next_item.bbox.y0 - item.bbox.y1 # If actual gap is less than desired spacing, add offset to push next item down if actual_gap < desired_spacing_after: additional_spacing = desired_spacing_after - actual_gap cumulative_y_offset -= additional_spacing # Negative because PDF Y increases upward logger.debug(f"Adding {additional_spacing:.1f}pt spacing after list item {item.element_id} " f"(actual_gap={actual_gap:.1f}pt, desired={desired_spacing_after:.1f}pt)") def _draw_text_with_spans( self, pdf_canvas: canvas.Canvas, spans: List['DocumentElement'], line_x: float, line_y: float, default_font_size: float, max_width: float = None ) -> float: """ Draw text with inline span styling (mixed styles within a line). Args: pdf_canvas: ReportLab canvas object spans: List of span DocumentElements line_x: Starting X position line_y: Y position default_font_size: Default font size if span has none max_width: Maximum width available (for scaling if needed) Returns: Total width of drawn text """ if not spans: return 0 # First pass: calculate total width with original sizes total_width = 0 span_data = [] # Store (span, text, font, size) for rendering for span in spans: span_text = span.get_text() if not span_text: continue # Apply span-specific styling to get font and size if span.style: self._apply_text_style(pdf_canvas, span.style, default_size=default_font_size) else: font_name = self.font_name if self.font_registered else 'Helvetica' pdf_canvas.setFont(font_name, default_font_size) current_font = pdf_canvas._fontname current_size = pdf_canvas._fontsize # Calculate span width span_width = pdf_canvas.stringWidth(span_text, current_font, current_size) total_width += span_width span_data.append((span, span_text, current_font, current_size, span_width)) # Calculate scale factor if needed scale_factor = 1.0 if max_width and total_width > max_width: scale_factor = (max_width / total_width) * 0.95 # 95% to leave margin logger.debug(f"Scaling spans: total_width={total_width:.1f}pt > max_width={max_width:.1f}pt, scale={scale_factor:.2f}") # Second pass: draw spans with scaling x_pos = line_x for span, span_text, font_name, original_size, span_width in span_data: # Apply scaled font size scaled_size = original_size * scale_factor scaled_size = max(scaled_size, 3) # Minimum 3pt # Set font with scaled size pdf_canvas.setFont(font_name, scaled_size) # Draw this span pdf_canvas.drawString(x_pos, line_y, span_text) # Calculate actual width with scaled size and advance position actual_width = pdf_canvas.stringWidth(span_text, font_name, scaled_size) x_pos += actual_width return total_width * scale_factor def _draw_text_element_direct( self, pdf_canvas: canvas.Canvas, element: 'DocumentElement', page_height: float, y_offset: float = 0 ): """ Draw text element with Direct track rich formatting. FIXED: Correctly handles multi-line blocks and spans coordinates. Prioritizes span-based rendering (using precise bbox from each span), falls back to block-level rendering with corrected Y-axis logic. Args: pdf_canvas: ReportLab canvas object element: DocumentElement with text content page_height: Page height for coordinate transformation y_offset: Optional Y coordinate offset (for list spacing), in PDF coordinates """ try: text_content = element.get_text() if not text_content: return # Get bounding box bbox = element.bbox if not bbox: logger.warning(f"No bbox for text element {element.element_id}") return bbox_width = bbox.x1 - bbox.x0 bbox_height = bbox.y1 - bbox.y0 # --- FIX 1: Prioritize Span-based Drawing (Precise Layout) --- # DirectExtractionEngine provides children (spans) with precise bboxes. # Using these preserves exact layout, kerning, and multi-column positioning. if element.children and len(element.children) > 0: for span in element.children: span_text = span.get_text() if not span_text: continue # Use span's own bbox for positioning s_bbox = span.bbox if not s_bbox: continue # Calculate font size from span style or bbox s_font_size = 10 # default if span.style and span.style.font_size: s_font_size = span.style.font_size else: # Estimate from bbox height s_font_size = (s_bbox.y1 - s_bbox.y0) * 0.75 s_font_size = max(min(s_font_size, 72), 4) # Apply span style if span.style: self._apply_text_style(pdf_canvas, span.style, default_size=s_font_size) else: font_name = self.font_name if self.font_registered else 'Helvetica' pdf_canvas.setFont(font_name, s_font_size) # Transform coordinates # PyMuPDF y1 is bottom of text box. ReportLab draws at baseline. # Using y1 with a small offset (20% of font size) approximates baseline position. span_pdf_x = s_bbox.x0 span_pdf_y = page_height - s_bbox.y1 + (s_font_size * 0.2) pdf_canvas.drawString(span_pdf_x, span_pdf_y + y_offset, span_text) # If we drew spans, we are done. Do not draw the block text on top. logger.debug(f"Drew {len(element.children)} spans using precise bbox positioning") return # --- FIX 2: Block-level Fallback (Corrected Y-Axis Logic) --- # Used when no spans are available (e.g. filtered text or modified structures) # Calculate font size from bbox height font_size = bbox_height * 0.75 font_size = max(min(font_size, 72), 4) # Clamp 4-72pt # Apply style if available alignment = 'left' # Default alignment if hasattr(element, 'style') and element.style: self._apply_text_style(pdf_canvas, element.style, default_size=font_size) # Get alignment from style if hasattr(element.style, 'alignment') and element.style.alignment: alignment = element.style.alignment else: # Use default font font_name = self.font_name if self.font_registered else 'Helvetica' pdf_canvas.setFont(font_name, font_size) # Detect list items and extract list properties is_list_item = (element.type == ElementType.LIST_ITEM) list_level = element.metadata.get('list_level', 0) if element.metadata else 0 # Get pre-computed list marker from metadata (set by _draw_list_elements_direct) list_marker = element.metadata.get('_list_marker', '') if element.metadata else '' list_type = element.metadata.get('_list_type') if element.metadata else None # If no pre-computed marker, remove original marker from text if is_list_item and list_marker: # Remove original marker from text content text_stripped = text_content.lstrip() # Remove ordered list marker text_content = re.sub(r'^\d+[\.\)]\s', '', text_stripped) # Remove unordered list marker text_content = re.sub(r'^[•·▪▫◦‣⁃]\s', '', text_content) # Get indentation from metadata (in points) indent = element.metadata.get('indent', 0) if element.metadata else 0 first_line_indent = element.metadata.get('first_line_indent', indent) if element.metadata else indent # Apply list indentation (20pt per level) if is_list_item: list_indent = list_level * 20 # 20pt per level indent += list_indent first_line_indent += list_indent # Get paragraph spacing paragraph_spacing_before = element.metadata.get('spacing_before', 0) if element.metadata else 0 paragraph_spacing_after = element.metadata.get('spacing_after', 0) if element.metadata else 0 # --- CRITICAL FIX: Start from TOP of block (y0), not bottom (y1) --- pdf_x = bbox.x0 pdf_y_top = page_height - bbox.y0 - paragraph_spacing_before + y_offset # Handle line breaks lines = text_content.split('\n') line_height = font_size * 1.2 # 120% of font size # Calculate list marker width for multi-line alignment marker_width = 0 if is_list_item and list_marker: # Use current font to calculate marker width marker_width = pdf_canvas.stringWidth(list_marker, pdf_canvas._fontname, font_size) # Draw each line with alignment for i, line in enumerate(lines): if not line.strip(): # Empty line: skip continue # Calculate Y position: Start from top, move down by line_height for each line # The first line's baseline is approx 1 line_height below the top line_y = pdf_y_top - ((i + 1) * line_height) + (font_size * 0.25) # 0.25 adjust for baseline # Get current font info font_name = pdf_canvas._fontname current_font_size = pdf_canvas._fontsize # Calculate line indentation line_indent = first_line_indent if i == 0 else indent # For list items: align subsequent lines with text after marker if is_list_item and i > 0 and marker_width > 0: line_indent += marker_width # Prepend list marker to first line rendered_line = line if is_list_item and i == 0 and list_marker: rendered_line = list_marker + line # Calculate text width text_width = pdf_canvas.stringWidth(rendered_line, font_name, current_font_size) available_width = bbox_width - line_indent # Scale font if needed if text_width > available_width and available_width > 0: scale_factor = available_width / text_width scaled_size = current_font_size * scale_factor * 0.95 scaled_size = max(scaled_size, 3) pdf_canvas.setFont(font_name, scaled_size) text_width = pdf_canvas.stringWidth(rendered_line, font_name, scaled_size) current_font_size = scaled_size # Calculate X position based on alignment line_x = pdf_x + line_indent if alignment == 'center': line_x = pdf_x + (bbox_width - text_width) / 2 elif alignment == 'right': line_x = pdf_x + bbox_width - text_width elif alignment == 'justify' and i < len(lines) - 1: # Justify: distribute extra space between words (except last line) words = rendered_line.split() if len(words) > 1: total_word_width = sum(pdf_canvas.stringWidth(word, font_name, current_font_size) for word in words) extra_space = available_width - total_word_width if extra_space > 0: word_spacing = extra_space / (len(words) - 1) # Draw words with calculated spacing x_pos = pdf_x + line_indent for word in words: pdf_canvas.drawString(x_pos, line_y, word) word_width = pdf_canvas.stringWidth(word, font_name, current_font_size) x_pos += word_width + word_spacing # Reset font for next line and skip normal drawString if text_width > available_width: pdf_canvas.setFont(font_name, font_size) continue # Draw the line at calculated position pdf_canvas.drawString(line_x, line_y, rendered_line) # Reset font size for next line if text_width > available_width: pdf_canvas.setFont(font_name, font_size) # Calculate actual text height used actual_text_height = len(lines) * line_height bbox_bottom_margin = bbox_height - actual_text_height - paragraph_spacing_before # Note: For list items, spacing_after is applied via y_offset in _draw_list_elements_direct # For other elements, spacing is inherent in element positioning (bbox-based layout) list_info = f", list={list_type}, level={list_level}" if is_list_item else "" y_offset_info = f", y_offset={y_offset:.1f}pt" if y_offset != 0 else "" logger.debug(f"Drew text element (fallback): {text_content[:30]}... " f"({len(lines)} lines, align={alignment}, indent={indent}{list_info}{y_offset_info}, " f"spacing_before={paragraph_spacing_before}, spacing_after={paragraph_spacing_after}, " f"actual_height={actual_text_height:.1f}, bbox_bottom_margin={bbox_bottom_margin:.1f})") except Exception as e: logger.error(f"Failed to draw text element {element.element_id}: {e}") def _build_rows_from_cells_dict(self, content: dict) -> list: """ Build row structure from cells dict (from Direct extraction JSON). The cells structure from Direct extraction: { "rows": 6, "cols": 2, "cells": [ {"row": 0, "col": 0, "content": "...", "row_span": 1, "col_span": 2}, {"row": 0, "col": 1, "content": "..."}, ... ] } Returns format compatible with HTMLTableParser output (with colspan/rowspan/col): [ {"cells": [{"text": "...", "colspan": 1, "rowspan": 1, "col": 0}, ...]}, {"cells": [{"text": "...", "colspan": 1, "rowspan": 1, "col": 0}, ...]}, ... ] Note: This returns actual cells per row with their absolute column positions. The table renderer uses 'col' to place cells correctly in the grid. """ try: num_rows = content.get('rows', 0) num_cols = content.get('cols', 0) cells = content.get('cells', []) if not cells or num_rows == 0 or num_cols == 0: return [] # Group cells by row cells_by_row = {} for cell in cells: row_idx = cell.get('row', 0) if row_idx not in cells_by_row: cells_by_row[row_idx] = [] cells_by_row[row_idx].append(cell) # Sort cells within each row by column for row_idx in cells_by_row: cells_by_row[row_idx].sort(key=lambda c: c.get('col', 0)) # Build rows structure with colspan/rowspan info and absolute col position rows_data = [] for row_idx in range(num_rows): row_cells = [] if row_idx in cells_by_row: for cell in cells_by_row[row_idx]: cell_content = cell.get('content', '') row_span = cell.get('row_span', 1) or 1 col_span = cell.get('col_span', 1) or 1 col_idx = cell.get('col', 0) row_cells.append({ 'text': str(cell_content) if cell_content else '', 'rowspan': row_span, 'colspan': col_span, 'col': col_idx # Absolute column position }) rows_data.append({'cells': row_cells}) logger.debug(f"Built {num_rows} rows from cells dict with span info") return rows_data except Exception as e: logger.error(f"Error building rows from cells dict: {e}") return [] def _draw_table_element_direct( self, pdf_canvas: canvas.Canvas, element: 'DocumentElement', page_height: float ): """ Draw table element with Direct track positioning. Args: pdf_canvas: ReportLab canvas object element: DocumentElement with table content page_height: Page height for coordinate transformation """ try: # Get table data - can be TableData object or dict from JSON rows_data = None if isinstance(element.content, TableData): # Direct TableData object - convert to HTML then parse html_content = element.content.to_html() parser = HTMLTableParser() parser.feed(html_content) if parser.tables and parser.tables[0]['rows']: rows_data = parser.tables[0]['rows'] elif isinstance(element.content, dict): # Dict from JSON - check if it has cells structure (from Direct extraction) if 'cells' in element.content: # Build rows from cells structure directly (avoid HTML round-trip) rows_data = self._build_rows_from_cells_dict(element.content) elif 'html' in element.content: # Has HTML content - parse it html_content = element.content['html'] parser = HTMLTableParser() parser.feed(html_content) if parser.tables and parser.tables[0]['rows']: rows_data = parser.tables[0]['rows'] if not rows_data: logger.warning(f"No table data for {element.element_id}") return rows = rows_data # Get bbox bbox = element.bbox if not bbox: logger.warning(f"No bbox for table {element.element_id}") return # Transform coordinates pdf_x = bbox.x0 # Use exact bbox position (no buffer) - scaling will ensure table fits pdf_y = page_height - bbox.y1 # Bottom of table (ReportLab Y coordinate) table_width = bbox.x1 - bbox.x0 table_height = bbox.y1 - bbox.y0 # Create table from reportlab.platypus import Table, TableStyle from reportlab.lib import colors # Determine grid size from rows structure # Note: rows may have 'col' attribute for absolute positioning (from Direct extraction) # or may be sequential (from HTML parsing) num_rows = len(rows) # Check if cells have absolute column positions has_absolute_cols = any( 'col' in cell for row in rows for cell in row['cells'] ) # Calculate actual number of columns max_cols = 0 if has_absolute_cols: # Use absolute col positions + colspan to find max column for row in rows: for cell in row['cells']: col = cell.get('col', 0) colspan = cell.get('colspan', 1) max_cols = max(max_cols, col + colspan) else: # Sequential cells: sum up colspans for row in rows: col_pos = 0 for cell in row['cells']: colspan = cell.get('colspan', 1) col_pos += colspan max_cols = max(max_cols, col_pos) # Build table data for ReportLab with proper grid structure # ReportLab needs a full grid with placeholders for spanned cells # and SPAN commands to merge them table_content = [] span_commands = [] covered = set() # Track cells covered by spans # First pass: mark covered cells and collect SPAN commands for row_idx, row in enumerate(rows): if has_absolute_cols: # Use absolute column positions for cell in row['cells']: col_pos = cell.get('col', 0) colspan = cell.get('colspan', 1) rowspan = cell.get('rowspan', 1) # Mark cells covered by this span if colspan > 1 or rowspan > 1: for r in range(row_idx, row_idx + rowspan): for c in range(col_pos, col_pos + colspan): if (r, c) != (row_idx, col_pos): covered.add((r, c)) # Add SPAN command for ReportLab span_commands.append(( 'SPAN', (col_pos, row_idx), (col_pos + colspan - 1, row_idx + rowspan - 1) )) else: # Sequential positioning col_pos = 0 for cell in row['cells']: while (row_idx, col_pos) in covered: col_pos += 1 colspan = cell.get('colspan', 1) rowspan = cell.get('rowspan', 1) if colspan > 1 or rowspan > 1: for r in range(row_idx, row_idx + rowspan): for c in range(col_pos, col_pos + colspan): if (r, c) != (row_idx, col_pos): covered.add((r, c)) span_commands.append(( 'SPAN', (col_pos, row_idx), (col_pos + colspan - 1, row_idx + rowspan - 1) )) col_pos += colspan # Second pass: build content grid for row_idx in range(num_rows): row_data = [''] * max_cols if row_idx < len(rows): if has_absolute_cols: # Place cells at their absolute positions for cell in rows[row_idx]['cells']: col_pos = cell.get('col', 0) if col_pos < max_cols: row_data[col_pos] = cell['text'].strip() else: # Sequential placement col_pos = 0 for cell in rows[row_idx]['cells']: while col_pos < max_cols and (row_idx, col_pos) in covered: col_pos += 1 if col_pos < max_cols: row_data[col_pos] = cell['text'].strip() colspan = cell.get('colspan', 1) col_pos += colspan table_content.append(row_data) logger.debug(f"Built table grid: {num_rows} rows × {max_cols} cols, {len(span_commands)} span commands (absolute_cols={has_absolute_cols})") # Use original column widths from extraction if available # Otherwise try to compute from cell_boxes (from PP-StructureV3) col_widths = None if element.metadata and 'column_widths' in element.metadata: col_widths = element.metadata['column_widths'] logger.debug(f"Using extracted column widths: {col_widths}") elif element.metadata and 'cell_boxes' in element.metadata: # Use cell_boxes from PP-StructureV3 for accurate column/row sizing cell_boxes = element.metadata['cell_boxes'] cell_boxes_source = element.metadata.get('cell_boxes_source', 'unknown') table_bbox_list = [bbox.x0, bbox.y0, bbox.x1, bbox.y1] logger.info(f"[TABLE] Using {len(cell_boxes)} cell boxes from {cell_boxes_source}") computed_col_widths, computed_row_heights = self._compute_table_grid_from_cell_boxes( cell_boxes, table_bbox_list, num_rows, max_cols ) if computed_col_widths: col_widths = computed_col_widths logger.info(f"[TABLE] Computed {len(col_widths)} column widths from cell_boxes") # NOTE: Don't use rowHeights from extraction - it causes content overlap # The extracted row heights are based on cell boundaries, not text content height. # When text wraps or uses different font sizes, the heights don't match. # Let ReportLab auto-calculate row heights based on content, then use scaling # to fit within the bbox (same approach as old commit ba8ddf2b). # Create table without rowHeights - let ReportLab auto-calculate t = Table(table_content, colWidths=col_widths) # Apply style with minimal padding to reduce table extension # Use Chinese font to support special characters (℃, μm, ≦, ×, Ω, etc.) font_for_table = self.font_name if self.font_registered else 'Helvetica' style_commands = [ ('GRID', (0, 0), (-1, -1), 0.5, colors.grey), ('FONTNAME', (0, 0), (-1, -1), font_for_table), ('FONTSIZE', (0, 0), (-1, -1), 8), ('ALIGN', (0, 0), (-1, -1), 'LEFT'), ('VALIGN', (0, 0), (-1, -1), 'TOP'), # Set minimal padding to prevent table from extending beyond bbox # User reported padding=1 was still insufficient ('TOPPADDING', (0, 0), (-1, -1), 0), ('BOTTOMPADDING', (0, 0), (-1, -1), 0), ('LEFTPADDING', (0, 0), (-1, -1), 1), ('RIGHTPADDING', (0, 0), (-1, -1), 1), ] # Add span commands for merged cells style_commands.extend(span_commands) if span_commands: logger.info(f"Applied {len(span_commands)} SPAN commands for merged cells") style = TableStyle(style_commands) t.setStyle(style) # Use canvas scaling as fallback to fit table within bbox # With proper row heights, scaling should be minimal (close to 1.0) # Step 1: Wrap to get actual rendered size actual_width, actual_height = t.wrapOn(pdf_canvas, table_width * 10, table_height * 10) logger.info(f"Table natural size: {actual_width:.1f} × {actual_height:.1f}pt, bbox: {table_width:.1f} × {table_height:.1f}pt") # Step 2: Calculate scale factor to fit within bbox scale_x = table_width / actual_width if actual_width > table_width else 1.0 scale_y = table_height / actual_height if actual_height > table_height else 1.0 scale = min(scale_x, scale_y, 1.0) # Never scale up, only down logger.info(f"Scale factor: {scale:.3f} (x={scale_x:.3f}, y={scale_y:.3f})") # Step 3: Draw with scaling using canvas transform pdf_canvas.saveState() pdf_canvas.translate(pdf_x, pdf_y) pdf_canvas.scale(scale, scale) t.drawOn(pdf_canvas, 0, 0) pdf_canvas.restoreState() logger.info(f"Drew table at ({pdf_x:.1f}, {pdf_y:.1f}) with scale {scale:.3f}, final size: {actual_width * scale:.1f} × {actual_height * scale:.1f}pt") logger.debug(f"Drew table element: {len(rows)} rows") except Exception as e: logger.error(f"Failed to draw table element {element.element_id}: {e}") def _draw_image_element_direct( self, pdf_canvas: canvas.Canvas, element: 'DocumentElement', page_height: float, result_dir: Path ): """ Draw image element with Direct track positioning. Args: pdf_canvas: ReportLab canvas object element: DocumentElement with image content page_height: Page height for coordinate transformation result_dir: Directory containing image files """ try: # Get image path image_path_str = self._get_image_path(element) if not image_path_str: logger.warning(f"No image path for element {element.element_id}") return # Construct full path to image # saved_path is relative to result_dir (e.g., "document_id_p1_img0.png") image_path = result_dir / image_path_str # Fallback for legacy data if not image_path.exists(): image_path = result_dir / Path(image_path_str).name if not image_path.exists(): logger.warning(f"Image not found: {image_path_str} (in {result_dir})") return # Get bbox bbox = element.bbox if not bbox: logger.warning(f"No bbox for image {element.element_id}") return # Transform coordinates pdf_x = bbox.x0 pdf_y = page_height - bbox.y1 # Bottom of image image_width = bbox.x1 - bbox.x0 image_height = bbox.y1 - bbox.y0 # Draw image pdf_canvas.drawImage( str(image_path), pdf_x, pdf_y, width=image_width, height=image_height, preserveAspectRatio=True ) logger.debug(f"Drew image: {image_path} (from: {image_path_str})") except Exception as e: logger.error(f"Failed to draw image element {element.element_id}: {e}") # ============================================================================ # Reflow Layout PDF Generation # ============================================================================ def _get_elements_in_reading_order(self, page_data: Dict) -> List[Dict]: """ Get elements sorted by reading order. For OCR track: Uses explicit 'reading_order' array from JSON For Direct track: Uses implicit element list order (PyMuPDF sort=True) Args: page_data: Page dictionary containing 'elements' and optionally 'reading_order' Returns: List of elements in proper reading order """ elements = page_data.get('elements', []) reading_order = page_data.get('reading_order') if reading_order and isinstance(reading_order, list): # OCR track: use explicit reading order ordered = [] for idx in reading_order: if isinstance(idx, int) and 0 <= idx < len(elements): ordered.append(elements[idx]) # Add any elements not in reading_order at the end ordered_indices = set(reading_order) for i, elem in enumerate(elements): if i not in ordered_indices: ordered.append(elem) return ordered else: # Direct track: elements already in reading order from PyMuPDF return elements def _get_reflow_styles(self) -> Dict[str, ParagraphStyle]: """Create consistent styles for reflow PDF generation.""" base_styles = getSampleStyleSheet() font_name = self.font_name if self.font_registered else 'Helvetica' styles = { 'Title': ParagraphStyle( 'ReflowTitle', parent=base_styles['Normal'], fontName=font_name, fontSize=18, leading=22, spaceAfter=12, textColor=colors.black, ), 'Heading1': ParagraphStyle( 'ReflowH1', parent=base_styles['Normal'], fontName=font_name, fontSize=16, leading=20, spaceAfter=10, spaceBefore=12, textColor=colors.black, ), 'Heading2': ParagraphStyle( 'ReflowH2', parent=base_styles['Normal'], fontName=font_name, fontSize=14, leading=18, spaceAfter=8, spaceBefore=10, textColor=colors.black, ), 'Body': ParagraphStyle( 'ReflowBody', parent=base_styles['Normal'], fontName=font_name, fontSize=12, leading=16, spaceAfter=6, textColor=colors.black, ), 'TableCell': ParagraphStyle( 'ReflowTableCell', parent=base_styles['Normal'], fontName=font_name, fontSize=10, leading=13, textColor=colors.black, ), 'Caption': ParagraphStyle( 'ReflowCaption', parent=base_styles['Normal'], fontName=font_name, fontSize=10, leading=13, spaceAfter=8, textColor=colors.gray, ), } return styles def _create_reflow_table(self, table_data: Dict, styles: Dict) -> Optional[Table]: """ Create a Platypus Table for reflow mode with merged cell support. Args: table_data: Table element dictionary with 'content' containing 'cells' styles: Style dictionary Returns: Platypus Table object or None """ try: # Get content - cells are inside 'content' dict content = table_data.get('content', {}) if not isinstance(content, dict): return None cells = content.get('cells', []) if not cells: return None # Determine grid dimensions num_rows = content.get('rows', 0) num_cols = content.get('cols', 0) if num_rows == 0 or num_cols == 0: # Calculate from cells for cell in cells: row = cell.get('row', cell.get('row_index', 0)) col = cell.get('col', cell.get('col_index', 0)) row_span = cell.get('row_span', 1) col_span = cell.get('col_span', 1) num_rows = max(num_rows, row + row_span) num_cols = max(num_cols, col + col_span) if num_rows == 0 or num_cols == 0: return None # Initialize grid with empty strings grid = [['' for _ in range(num_cols)] for _ in range(num_rows)] # Track which cells are covered by spans covered = [[False for _ in range(num_cols)] for _ in range(num_rows)] # Track span commands span_commands = [] # Fill grid with cell content for cell in cells: row = cell.get('row', cell.get('row_index', 0)) col = cell.get('col', cell.get('col_index', 0)) row_span = cell.get('row_span', 1) col_span = cell.get('col_span', 1) # Get cell text text = cell.get('content', cell.get('text', '')) if not isinstance(text, str): text = str(text) if text else '' # Escape HTML special characters text = text.replace('&', '&').replace('<', '<').replace('>', '>') # Place content in the top-left cell of the span if 0 <= row < num_rows and 0 <= col < num_cols: grid[row][col] = text # Mark covered cells for spans if row_span > 1 or col_span > 1: # Add SPAN command span_commands.append(( 'SPAN', (col, row), (col + col_span - 1, row + row_span - 1) )) # Mark cells as covered for r in range(row, min(row + row_span, num_rows)): for c in range(col, min(col + col_span, num_cols)): if r != row or c != col: covered[r][c] = True # Build table data with Paragraphs data = [] for row_idx in range(num_rows): row_data = [] for col_idx in range(num_cols): if covered[row_idx][col_idx]: # Empty cell for covered spans row_data.append('') else: text = grid[row_idx][col_idx] row_data.append(Paragraph(text, styles['TableCell'])) data.append(row_data) if not data: return None # Create table table = Table(data) # Build style commands style_commands = [ ('GRID', (0, 0), (-1, -1), 0.5, colors.black), ('VALIGN', (0, 0), (-1, -1), 'TOP'), ('LEFTPADDING', (0, 0), (-1, -1), 6), ('RIGHTPADDING', (0, 0), (-1, -1), 6), ('TOPPADDING', (0, 0), (-1, -1), 4), ('BOTTOMPADDING', (0, 0), (-1, -1), 4), ('BACKGROUND', (0, 0), (-1, 0), colors.lightgrey), # Header row ] # Add span commands style_commands.extend(span_commands) table.setStyle(TableStyle(style_commands)) logger.info(f"[REFLOW TABLE] Created table with {num_rows}x{num_cols} cells, {len(span_commands)} spans") return table except Exception as e: logger.error(f"Failed to create reflow table: {e}") import traceback traceback.print_exc() return None def _embed_image_reflow( self, element: Dict, result_dir: Path, max_width: float = 450 ) -> Optional[PlatypusImage]: """ Embed an image for reflow mode. Args: element: Image element dictionary result_dir: Directory containing images max_width: Maximum width in points Returns: Platypus Image object or None """ try: # Get image path - check multiple possible locations img_path_str = element.get('image_path', element.get('path', '')) # Also check content.saved_path (Direct track format) if not img_path_str: content = element.get('content', {}) if isinstance(content, dict): img_path_str = content.get('saved_path', content.get('path', '')) if not img_path_str: return None img_path = result_dir / img_path_str if not img_path.exists(): # Try just the filename img_path = result_dir / Path(img_path_str).name if not img_path.exists(): logger.warning(f"Image not found for reflow: {img_path_str}") return None # Create Platypus Image img = PlatypusImage(str(img_path)) # Scale to fit page width if necessary if img.drawWidth > max_width: ratio = max_width / img.drawWidth img.drawWidth = max_width img.drawHeight *= ratio return img except Exception as e: logger.error(f"Failed to embed image for reflow: {e}") return None def generate_reflow_pdf( self, json_path: Path, output_path: Path, source_file_path: Optional[Path] = None ) -> bool: """ Generate reflow layout PDF from OCR/Direct JSON data. This creates a flowing document with consistent font sizes, proper reading order, and inline tables/images. Args: json_path: Path to result JSON file (UnifiedDocument format) output_path: Path to save generated PDF source_file_path: Optional path to original source file (for images) Returns: True if successful, False otherwise """ try: # Load JSON data logger.info(f"Generating reflow PDF from: {json_path}") with open(json_path, 'r', encoding='utf-8') as f: json_data = json.load(f) # Get styles styles = self._get_reflow_styles() # Build document content story = [] # Use source_file_path if provided (for translated PDFs where JSON is in temp dir) # Otherwise use json_path.parent (for regular reflow PDFs) if source_file_path and source_file_path.is_dir(): result_dir = source_file_path elif source_file_path and source_file_path.is_file(): result_dir = source_file_path.parent else: result_dir = json_path.parent # Process each page pages = json_data.get('pages', []) for page_idx, page_data in enumerate(pages): if page_idx > 0: # Add page break between pages story.append(Spacer(1, 30)) # Get elements in reading order elements = self._get_elements_in_reading_order(page_data) for elem in elements: elem_type = elem.get('type', elem.get('element_type', 'text')) content = elem.get('content', elem.get('text', '')) # Types that can have dict content (handled specially) dict_content_types = ('table', 'Table', 'image', 'figure', 'Image', 'Figure', 'chart', 'Chart') # Ensure content is a string for text elements if isinstance(content, dict): # Tables, images, charts have dict content - handled by their respective methods if elem_type not in dict_content_types: # Skip other elements with dict content continue elif not isinstance(content, str): content = str(content) if content else '' if elem_type in ('table', 'Table'): # Handle table table = self._create_reflow_table(elem, styles) if table: story.append(table) story.append(Spacer(1, 12)) # Handle embedded images in table (from metadata) metadata = elem.get('metadata', {}) embedded_images = metadata.get('embedded_images', []) for emb_img in embedded_images: img_path_str = emb_img.get('saved_path', '') if img_path_str: img_path = result_dir / img_path_str if not img_path.exists(): img_path = result_dir / Path(img_path_str).name if img_path.exists(): try: img = PlatypusImage(str(img_path)) # Scale to fit page width if necessary max_width = 450 if img.drawWidth > max_width: ratio = max_width / img.drawWidth img.drawWidth = max_width img.drawHeight *= ratio story.append(img) story.append(Spacer(1, 8)) logger.info(f"Embedded table image in reflow: {img_path.name}") except Exception as e: logger.warning(f"Failed to embed table image: {e}") elif elem_type in ('image', 'figure', 'Image', 'Figure', 'chart', 'Chart'): # Handle image/chart img = self._embed_image_reflow(elem, result_dir) if img: story.append(img) story.append(Spacer(1, 8)) elif elem_type in ('title', 'Title'): # Title text if content: content = content.replace('&', '&').replace('<', '<').replace('>', '>') story.append(Paragraph(content, styles['Title'])) elif elem_type in ('section_header', 'SectionHeader', 'h1', 'H1'): # Heading 1 if content: content = content.replace('&', '&').replace('<', '<').replace('>', '>') story.append(Paragraph(content, styles['Heading1'])) elif elem_type in ('h2', 'H2', 'Heading2'): # Heading 2 if content: content = content.replace('&', '&').replace('<', '<').replace('>', '>') story.append(Paragraph(content, styles['Heading2'])) else: # Body text (default) if content: content = content.replace('&', '&').replace('<', '<').replace('>', '>') story.append(Paragraph(content, styles['Body'])) if not story: logger.warning("No content to generate reflow PDF") return False # Create PDF document doc = SimpleDocTemplate( str(output_path), pagesize=A4, leftMargin=50, rightMargin=50, topMargin=50, bottomMargin=50 ) # Build PDF doc.build(story) logger.info(f"Generated reflow PDF: {output_path} ({output_path.stat().st_size} bytes)") return True except Exception as e: logger.error(f"Failed to generate reflow PDF: {e}") import traceback traceback.print_exc() return False def generate_translated_pdf( self, result_json_path: Path, translation_json_path: Path, output_path: Path, source_file_path: Optional[Path] = None ) -> bool: """ Generate reflow layout PDF with translated content. This method loads the original result JSON and translation JSON, merges them to replace original content with translations, and generates a PDF with the translated content at original positions. Args: result_json_path: Path to original result JSON file (UnifiedDocument format) translation_json_path: Path to translation JSON file output_path: Path to save generated translated PDF source_file_path: Optional path to original source file Returns: True if successful, False otherwise """ import tempfile try: # Import apply_translations from translation service from app.services.translation_service import apply_translations # Load original result JSON logger.info(f"Loading result JSON: {result_json_path}") with open(result_json_path, 'r', encoding='utf-8') as f: result_json = json.load(f) # Load translation JSON logger.info(f"Loading translation JSON: {translation_json_path}") with open(translation_json_path, 'r', encoding='utf-8') as f: translation_json = json.load(f) # Extract translations dict from translation JSON translations = translation_json.get('translations', {}) if not translations: logger.warning("No translations found in translation JSON") # Still generate PDF with original content as fallback return self.generate_layout_pdf( json_path=result_json_path, output_path=output_path, source_file_path=source_file_path ) # Apply translations to result JSON translated_doc = apply_translations(result_json, translations) target_lang = translation_json.get('target_lang', 'unknown') logger.info( f"Generating translated PDF: {len(translations)} translations applied, " f"target_lang={target_lang}" ) # Write translated JSON to a temporary file and use reflow PDF generation with tempfile.NamedTemporaryFile( mode='w', suffix='_translated.json', delete=False, encoding='utf-8' ) as tmp_file: json.dump(translated_doc, tmp_file, ensure_ascii=False, indent=2) tmp_path = Path(tmp_file.name) try: # Use reflow PDF generation for better translated content display # Pass result_json_path.parent as image directory (not the temp file's parent) success = self.generate_reflow_pdf( json_path=tmp_path, output_path=output_path, source_file_path=result_json_path.parent # Contains extracted images ) return success finally: # Clean up temporary file if tmp_path.exists(): tmp_path.unlink() except FileNotFoundError as e: logger.error(f"File not found: {e}") return False except json.JSONDecodeError as e: logger.error(f"Invalid JSON: {e}") return False except Exception as e: logger.error(f"Failed to generate translated PDF: {e}") import traceback traceback.print_exc() return False def generate_translated_layout_pdf( self, result_json_path: Path, translation_json_path: Path, output_path: Path, source_file_path: Optional[Path] = None ) -> bool: """ Generate layout-preserving PDF with translated content. This method creates a PDF that maintains the original document layout while displaying translated text. Key features: - Text wraps within original bounding boxes (no font shrinking) - Tables adapt to translated content - Images and other elements remain at original positions - Font size is kept readable (minimum 10pt) Args: result_json_path: Path to original result JSON file translation_json_path: Path to translation JSON file output_path: Path to save generated translated PDF source_file_path: Optional path for image directory Returns: True if successful, False otherwise """ import tempfile from reportlab.lib.styles import getSampleStyleSheet, ParagraphStyle from reportlab.platypus import Paragraph, Frame from reportlab.lib.enums import TA_LEFT, TA_CENTER, TA_RIGHT try: # Import apply_translations from translation service from app.services.translation_service import apply_translations # Load original result JSON logger.info(f"Loading result JSON for layout PDF: {result_json_path}") with open(result_json_path, 'r', encoding='utf-8') as f: result_json = json.load(f) # Load translation JSON logger.info(f"Loading translation JSON: {translation_json_path}") with open(translation_json_path, 'r', encoding='utf-8') as f: translation_json = json.load(f) # Extract translations dict translations = translation_json.get('translations', {}) if not translations: logger.warning("No translations found, falling back to original layout PDF") return self.generate_layout_pdf( json_path=result_json_path, output_path=output_path, source_file_path=source_file_path ) # Apply translations to result JSON translated_doc = apply_translations(result_json, translations) target_lang = translation_json.get('target_lang', 'unknown') logger.info( f"Generating translated layout PDF: {len(translations)} translations, " f"target_lang={target_lang}" ) # Determine image directory if source_file_path and source_file_path.is_dir(): image_dir = source_file_path elif source_file_path and source_file_path.is_file(): image_dir = source_file_path.parent else: image_dir = result_json_path.parent # Create PDF canvas from reportlab.pdfgen import canvas # Get page dimensions from first page pages = translated_doc.get('pages', []) if not pages: logger.error("No pages in document") return False first_page = pages[0] dims = first_page.get('dimensions', {}) page_width = dims.get('width', 595.32) page_height = dims.get('height', 841.92) pdf_canvas = canvas.Canvas(str(output_path), pagesize=(page_width, page_height)) # Create paragraph styles for text wrapping base_style = ParagraphStyle( 'TranslatedBase', fontName=self.font_name if self.font_registered else 'Helvetica', fontSize=10, leading=12, wordWrap='CJK', # Support CJK word wrapping ) # Process each page for page_idx, page_data in enumerate(pages): logger.info(f"Processing translated layout page {page_idx + 1}/{len(pages)}") # Get current page dimensions dims = page_data.get('dimensions', {}) current_page_width = dims.get('width', page_width) current_page_height = dims.get('height', page_height) if page_idx > 0: pdf_canvas.showPage() pdf_canvas.setPageSize((current_page_width, current_page_height)) # Process elements: # - Tables: draw borders + translated cell text (with dynamic font sizing) # - Text elements: draw at original positions, SKIP if inside table bbox # - Images: draw at original positions elements = page_data.get('elements', []) # Collect table bboxes to skip text elements inside tables table_bboxes = [] for elem in elements: if elem.get('type') in ('table', 'Table'): elem_bbox = elem.get('bbox', {}) if elem_bbox: table_bboxes.append(elem_bbox) def is_inside_table(text_bbox): """Check if text bbox is inside any table bbox.""" margin = 5 for tb in table_bboxes: if (text_bbox.get('x0', 0) >= tb.get('x0', 0) - margin and text_bbox.get('y0', 0) >= tb.get('y0', 0) - margin and text_bbox.get('x1', 0) <= tb.get('x1', 0) + margin and text_bbox.get('y1', 0) <= tb.get('y1', 0) + margin): return True return False for elem in elements: elem_type = elem.get('type', 'text') content = elem.get('content', '') bbox = elem.get('bbox', {}) if not bbox: continue x0 = bbox.get('x0', 0) y0 = bbox.get('y0', 0) x1 = bbox.get('x1', 0) y1 = bbox.get('y1', 0) box_width = x1 - x0 box_height = y1 - y0 if box_width <= 0 or box_height <= 0: continue # Handle different element types if elem_type in ('image', 'figure', 'Image', 'Figure', 'chart', 'Chart'): # Skip large vector_graphics charts - they're visual decorations that cover text if elem_type in ('chart', 'Chart'): elem_content = elem.get('content', {}) is_vector_graphics = ( isinstance(elem_content, dict) and elem_content.get('source') == 'vector_graphics' ) if is_vector_graphics: page_area = current_page_width * current_page_height elem_area = box_width * box_height coverage_ratio = elem_area / page_area if page_area > 0 else 0 if coverage_ratio > 0.5: logger.info(f"Skipping large vector_graphics chart " f"(covers {coverage_ratio*100:.1f}% of page)") continue # Draw image img = self._embed_image_reflow(elem, image_dir) if img: # Convert to PDF coordinates pdf_y = current_page_height - y1 # Scale image to fit bbox scale = min(box_width / img.drawWidth, box_height / img.drawHeight) img.drawWidth *= scale img.drawHeight *= scale img.drawOn(pdf_canvas, x0, pdf_y) elif elem_type in ('table', 'Table'): # Draw table with wrapping self._draw_translated_table( pdf_canvas, elem, current_page_height, image_dir ) elif isinstance(content, str) and content.strip(): # Skip text elements inside table bboxes # (Table cells are rendered by _draw_translated_table with dynamic font sizing) if is_inside_table(bbox): continue # Text element - use Paragraph for word wrapping # Escape special characters safe_content = content.replace('&', '&') safe_content = safe_content.replace('<', '<') safe_content = safe_content.replace('>', '>') # Replace newlines with
safe_content = safe_content.replace('\n', '
') # Get original font size from style info style_info = elem.get('style', {}) original_font_size = style_info.get('font_size', 12.0) # Detect vertical text (Y-axis labels, etc.) # Vertical text has aspect_ratio (height/width) > 2 and multiple characters is_vertical_text = ( box_height > box_width * 2 and len(content.strip()) > 1 ) if is_vertical_text: # For vertical text, use original font size and rotate font_size = min(original_font_size, box_width * 0.9) font_size = max(font_size, 6) # Minimum 6pt # Save canvas state for rotation pdf_canvas.saveState() # Convert to PDF coordinates pdf_y_center = current_page_height - (y0 + y1) / 2 x_center = (x0 + x1) / 2 # Translate to center, rotate, translate back pdf_canvas.translate(x_center, pdf_y_center) pdf_canvas.rotate(90) # Set font and draw text centered pdf_canvas.setFont( self.font_name if self.font_registered else 'Helvetica', font_size ) # Draw text at origin (since we translated to center) text_width = pdf_canvas.stringWidth( safe_content.replace('&', '&').replace('<', '<').replace('>', '>'), self.font_name if self.font_registered else 'Helvetica', font_size ) pdf_canvas.drawString(-text_width / 2, -font_size / 3, safe_content.replace('&', '&').replace('<', '<').replace('>', '>')) pdf_canvas.restoreState() else: # For horizontal text, dynamically fit text within bbox # Start with original font size and reduce until text fits MIN_FONT_SIZE = 6 MAX_FONT_SIZE = 14 if original_font_size > 0: start_font_size = min(original_font_size, MAX_FONT_SIZE) else: start_font_size = min(box_height * 0.7, MAX_FONT_SIZE) font_size = max(start_font_size, MIN_FONT_SIZE) # Try progressively smaller font sizes until text fits para = None para_height = box_height + 1 # Start with height > box to enter loop while font_size >= MIN_FONT_SIZE and para_height > box_height: elem_style = ParagraphStyle( f'elem_{id(elem)}_{font_size}', parent=base_style, fontSize=font_size, leading=font_size * 1.15, # Tighter leading ) para = Paragraph(safe_content, elem_style) para_width, para_height = para.wrap(box_width, box_height * 3) if para_height <= box_height: break # Text fits! font_size -= 0.5 # Reduce font size and try again # Ensure minimum font size if font_size < MIN_FONT_SIZE: font_size = MIN_FONT_SIZE elem_style = ParagraphStyle( f'elem_{id(elem)}_min', parent=base_style, fontSize=font_size, leading=font_size * 1.15, ) para = Paragraph(safe_content, elem_style) para_width, para_height = para.wrap(box_width, box_height * 3) # Convert to PDF coordinates (y from bottom) # Clip to bbox height to prevent overflow actual_height = min(para_height, box_height) pdf_y = current_page_height - y0 - actual_height # Draw the paragraph para.drawOn(pdf_canvas, x0, pdf_y) # Save PDF pdf_canvas.save() logger.info(f"Translated layout PDF saved to {output_path}") return True except FileNotFoundError as e: logger.error(f"File not found: {e}") return False except json.JSONDecodeError as e: logger.error(f"Invalid JSON: {e}") return False except Exception as e: logger.error(f"Failed to generate translated layout PDF: {e}") import traceback traceback.print_exc() return False def _draw_translated_table( self, pdf_canvas, elem: Dict, page_height: float, image_dir: Path ): """ Draw a table with translated content. Approach: 1. Draw cell borders using cell_boxes from metadata 2. Render translated text in each cell with dynamic font sizing 3. Draw embedded images at their original positions Text is rendered with dynamic font sizing to fit within cells. Minimum font size is 6pt for readability. Args: pdf_canvas: ReportLab canvas elem: Table element dict with metadata containing cell_boxes page_height: Page height for coordinate transformation image_dir: Directory containing images """ from reportlab.lib import colors from reportlab.lib.styles import ParagraphStyle from reportlab.platypus import Paragraph MIN_FONT_SIZE = 6 # Minimum font size for readability try: content = elem.get('content', {}) bbox = elem.get('bbox', {}) metadata = elem.get('metadata', {}) if not bbox: return # Get table bounding box if isinstance(bbox, dict): tx0 = bbox.get('x0', 0) ty0 = bbox.get('y0', 0) tx1 = bbox.get('x1', 0) ty1 = bbox.get('y1', 0) else: tx0, ty0, tx1, ty1 = bbox[:4] if len(bbox) >= 4 else (0, 0, 0, 0) table_width = tx1 - tx0 table_height = ty1 - ty0 # Step 1: Draw outer table border pdf_canvas.setStrokeColor(colors.black) pdf_canvas.setLineWidth(1.0) pdf_y_bottom = page_height - ty1 pdf_canvas.rect(tx0, pdf_y_bottom, table_width, table_height, stroke=1, fill=0) # Step 2: Get or calculate cell boxes cell_boxes = metadata.get('cell_boxes', []) # If no cell_boxes, calculate from column_widths and row_heights if not cell_boxes: column_widths = metadata.get('column_widths', []) row_heights = metadata.get('row_heights', []) if column_widths and row_heights: # Calculate cell positions from widths and heights cell_boxes = [] rows = content.get('rows', len(row_heights)) if isinstance(content, dict) else len(row_heights) cols = content.get('cols', len(column_widths)) if isinstance(content, dict) else len(column_widths) # Calculate cumulative positions x_positions = [tx0] for w in column_widths[:cols]: x_positions.append(x_positions[-1] + w) y_positions = [ty0] for h in row_heights[:rows]: y_positions.append(y_positions[-1] + h) # Create cell boxes for each cell (row-major order) for row_idx in range(rows): for col_idx in range(cols): if col_idx < len(x_positions) - 1 and row_idx < len(y_positions) - 1: cx0 = x_positions[col_idx] cy0 = y_positions[row_idx] cx1 = x_positions[col_idx + 1] cy1 = y_positions[row_idx + 1] cell_boxes.append([cx0, cy0, cx1, cy1]) logger.debug(f"Calculated {len(cell_boxes)} cell boxes from {cols} cols x {rows} rows") # Normalize cell boxes for grid alignment if cell_boxes and hasattr(self, '_normalize_cell_boxes_to_grid'): cell_boxes = self._normalize_cell_boxes_to_grid(cell_boxes) # Draw cell borders if cell_boxes: pdf_canvas.setLineWidth(0.5) for box in cell_boxes: if len(box) >= 4: cx0, cy0, cx1, cy1 = box[:4] cell_width = cx1 - cx0 cell_height = cy1 - cy0 pdf_cell_y = page_height - cy1 pdf_canvas.rect(cx0, pdf_cell_y, cell_width, cell_height, stroke=1, fill=0) # Step 3: Render translated text in each cell cells = content.get('cells', []) if isinstance(content, dict) else [] font_name = self.font_name if self.font_registered else 'Helvetica' for i, cell in enumerate(cells): cell_text = cell.get('content', cell.get('text', '')) if not cell_text or not cell_text.strip(): continue # Get cell bounding box by index if i >= len(cell_boxes): continue cx0, cy0, cx1, cy1 = cell_boxes[i][:4] cell_width = cx1 - cx0 cell_height = cy1 - cy0 # Skip tiny cells if cell_width < 10 or cell_height < 10: continue # Prepare text (escape HTML special chars) safe_text = str(cell_text).replace('&', '&') safe_text = safe_text.replace('<', '<').replace('>', '>') safe_text = safe_text.replace('\n', '
') # Dynamic font sizing: start at 10pt, shrink until text fits padding = 3 available_width = cell_width - padding * 2 available_height = cell_height - padding * 2 if available_width <= 0 or available_height <= 0: continue # Try font sizes from 10pt down to MIN_FONT_SIZE for font_size in range(10, MIN_FONT_SIZE - 1, -1): cell_style = ParagraphStyle( f'cell_{i}_{font_size}', fontName=font_name, fontSize=font_size, leading=font_size * 1.15, wordWrap='CJK', ) para = Paragraph(safe_text, cell_style) para_width, para_height = para.wrap(available_width, available_height * 10) if para_height <= available_height: break # Text fits at this font size # Draw text (centered vertically in cell) text_x = cx0 + padding # Calculate vertical position (top-aligned within cell) text_y = page_height - cy0 - padding - min(para_height, available_height) para.drawOn(pdf_canvas, text_x, text_y) logger.info(f"[TRANSLATED TABLE] Drew table with {len(cell_boxes)} borders, {len(cells)} cells") # Step 4: Draw embedded images embedded_images = metadata.get('embedded_images', []) if embedded_images and image_dir: for emb_img in embedded_images: self._draw_embedded_image( pdf_canvas, emb_img, page_height, image_dir, 1.0, 1.0 ) except Exception as e: logger.error(f"Failed to draw translated table: {e}") import traceback traceback.print_exc() # Singleton instance pdf_generator_service = PDFGeneratorService()