xgen-doc2chunk 0.1.0__py3-none-any.whl → 0.1.2__py3-none-any.whl

xgen_doc2chunk/core/processor/pdf_helpers/pdf_table_processor.py

@@ -0,0 +1,897 @@
+# xgen_doc2chunk/core/processor/pdf_helpers/pdf_table_processor.py
+"""
+PDF Table Processing Module
+
+Provides functions for table extraction, merging, annotation integration,
+and HTML conversion from PDF documents.
+"""
+import copy
+import logging
+from typing import Any, Dict, List, Optional, Tuple, Set
+from dataclasses import dataclass
+from collections import defaultdict
+
+from xgen_doc2chunk.core.processor.pdf_helpers.types import (
+    TableDetectionStrategy,
+    ElementType,
+    PDFConfig,
+    PageElement,
+    PageBorderInfo,
+    CellInfo,
+)
+from xgen_doc2chunk.core.processor.pdf_helpers.pdf_utils import (
+    escape_html,
+    get_text_lines_with_positions,
+)
+from xgen_doc2chunk.core.processor.pdf_helpers.pdf_table_detection import TableDetectionEngine
+from xgen_doc2chunk.core.processor.pdf_helpers.pdf_cell_analysis import CellAnalysisEngine
+
+logger = logging.getLogger("document-processor")
+
+
+# ============================================================================
+# Data Classes
+# ============================================================================
+
+@dataclass
+class AnnotationInfo:
+    """Annotation/footnote/endnote info."""
+    text: str
+    bbox: Tuple[float, float, float, float]
+    type: str  # 'footnote', 'endnote', 'table_note'
+    related_table_idx: Optional[int] = None
+
+
+@dataclass
+class TableInfo:
+    """Final table info."""
+    page_num: int
+    table_idx: int
+    bbox: Tuple[float, float, float, float]
+    data: List[List[Optional[str]]]
+    col_count: int
+    row_count: int
+    page_height: float
+    cells_info: Optional[List[Dict]] = None
+    annotations: Optional[List[AnnotationInfo]] = None
+    detection_strategy: Optional[TableDetectionStrategy] = None
+    confidence: float = 1.0
+
+
+# ============================================================================
+# Table Extraction
+# ============================================================================
+
+def extract_all_tables(
+    doc,
+    file_path: str,
+    detect_page_border_func,
+    is_table_likely_border_func
+) -> Dict[int, List[PageElement]]:
+    """
+    Extracts tables from entire document.
+
+    Strategy:
+    1. Multi-strategy table detection
+    2. Select best result based on confidence
+    3. Cell analysis and merge cell processing
+    4. Annotation integration
+    5. Cross-page continuity handling
+
+    Args:
+        doc: PyMuPDF document object
+        file_path: PDF file path
+        detect_page_border_func: Function to detect page borders
+        is_table_likely_border_func: Function to check if table is a border
+
+    Returns:
+        Dictionary mapping page numbers to list of table PageElements
+    """
+    tables_by_page: Dict[int, List[PageElement]] = {}
+    all_table_infos: List[TableInfo] = []
+
+    # Step 1: Detect tables on each page
+    for page_num in range(len(doc)):
+        page = doc[page_num]
+        page_height = page.rect.height
+
+        # Detect page border
+        border_info = detect_page_border_func(page)
+
+        try:
+            # Use table detection engine
+            detection_engine = TableDetectionEngine(page, page_num, file_path)
+            candidates = detection_engine.detect_tables()
+
+            for idx, candidate in enumerate(candidates):
+                # Check if overlaps with page border
+                if border_info.has_border and is_table_likely_border_func(
+                    candidate.bbox, border_info, page
+                ):
+                    logger.debug(f"[PDF] Skipping page border table: {candidate.bbox}")
+                    continue
+
+                # Convert cell info to dictionary
+                cells_info = None
+                if candidate.cells:
+                    cells_info = [
+                        {
+                            'row': cell.row,
+                            'col': cell.col,
+                            'rowspan': cell.rowspan,
+                            'colspan': cell.colspan,
+                            'bbox': cell.bbox
+                        }
+                        for cell in candidate.cells
+                    ]
+
+                table_info = TableInfo(
+                    page_num=page_num,
+                    table_idx=idx,
+                    bbox=candidate.bbox,
+                    data=candidate.data,
+                    col_count=candidate.col_count,
+                    row_count=candidate.row_count,
+                    page_height=page_height,
+                    cells_info=cells_info,
+                    detection_strategy=candidate.strategy,
+                    confidence=candidate.confidence
+                )
+
+                all_table_infos.append(table_info)
+
+        except Exception as e:
+            logger.debug(f"[PDF] Error detecting tables on page {page_num}: {e}")
+            continue
+
+    # Step 2: Merge adjacent tables
+    merged_tables = merge_adjacent_tables(all_table_infos)
+
+    # Step 3: Find and insert annotations
+    merged_tables = find_and_insert_annotations(doc, merged_tables)
+
+    # Step 4: Handle table continuity
+    processed_tables = process_table_continuity(merged_tables)
+
+    # Step 5: HTML conversion and PageElement creation
+    # Single-column tables as TEXT, 2+ columns as TABLE
+    single_col_count = 0
+    real_table_count = 0
+
+    for table_info in processed_tables:
+        try:
+            page_num = table_info.page_num
+
+            if page_num not in tables_by_page:
+                tables_by_page[page_num] = []
+
+            # Check if single-column table
+            if is_single_column_table(table_info):
+                # Single-column table: convert to text list as TEXT type
+                text_content = convert_single_column_to_text(table_info)
+
+                if text_content:
+                    tables_by_page[page_num].append(PageElement(
+                        element_type=ElementType.TEXT,
+                        content=text_content,
+                        bbox=table_info.bbox,
+                        page_num=page_num
+                    ))
+                    single_col_count += 1
+            else:
+                # 2+ columns: convert to HTML table
+                html_table = convert_table_to_html(table_info)
+
+                if html_table:
+                    tables_by_page[page_num].append(PageElement(
+                        element_type=ElementType.TABLE,
+                        content=html_table,
+                        bbox=table_info.bbox,
+                        page_num=page_num
+                    ))
+                    real_table_count += 1
+
+        except Exception as e:
+            logger.debug(f"[PDF] Error converting table to HTML: {e}")
+            continue
+
+    if single_col_count > 0:
+        logger.info(f"[PDF] Converted {single_col_count} single-column tables to text")
+    logger.info(f"[PDF] Extracted {real_table_count} tables from {len(tables_by_page)} pages")
+    return tables_by_page
+
+
+# ============================================================================
+# Annotation Integration
+# ============================================================================
+
+def find_and_insert_annotations(doc, tables: List[TableInfo]) -> List[TableInfo]:
+    """
+    Finds and integrates annotations/footnotes/endnotes inside and after tables.
+
+    Detection patterns:
+    1. Rows starting with "Note)" etc. right after table
+    2. Subheader rows inside table (e.g., (A), (B))
+    3. Footnote/endnote markers (?? *, ?? ?? etc.)
+
+    Args:
+        doc: PyMuPDF document object
+        tables: List of TableInfo
+
+    Returns:
+        Updated list of TableInfo with annotations
+    """
+    if not tables:
+        return tables
+
+    result = []
+    tables_by_page: Dict[int, List[TableInfo]] = defaultdict(list)
+
+    for table in tables:
+        tables_by_page[table.page_num].append(table)
+
+    for page_num, page_tables in tables_by_page.items():
+        page = doc[page_num]
+        page_height = page.rect.height
+
+        sorted_tables = sorted(page_tables, key=lambda t: t.bbox[1])
+        text_lines = get_text_lines_with_positions(page)
+
+        for i, table in enumerate(sorted_tables):
+            table_top = table.bbox[1]
+            table_bottom = table.bbox[3]
+            table_left = table.bbox[0]
+            table_right = table.bbox[2]
+
+            next_table_top = sorted_tables[i + 1].bbox[1] if i + 1 < len(sorted_tables) else page_height
+
+            # 1. Find annotation rows right after table
+            annotation_lines = []
+            for line in text_lines:
+                # Right below table, before next table
+                if table_bottom - 3 <= line['y0'] <= table_bottom + PDFConfig.ANNOTATION_Y_MARGIN:
+                    if line['x0'] >= table_left - 10 and line['x1'] <= table_right + 10:
+                        if line['y0'] < next_table_top - 20:
+                            # Check annotation pattern
+                            for pattern in PDFConfig.ANNOTATION_PATTERNS:
+                                if line['text'].startswith(pattern):
+                                    annotation_lines.append(line)
+                                    break
+
+            if annotation_lines:
+                table = add_annotation_to_table(table, annotation_lines, 'footer')
+                logger.debug(f"[PDF] Added annotation to table on page {page_num + 1}")
+
+            # 2. Find subheader rows (e.g., (A), (B)) - only when no subheader exists
+            has_subheader = False
+            if table.row_count >= 2 and table.data and len(table.data) >= 2:
+                # Check if second row is subheader pattern
+                second_row = table.data[1] if len(table.data) > 1 else []
+                for cell in second_row:
+                    if cell and ('(A)' in str(cell) or '(B)' in str(cell)):
+                        has_subheader = True
+                        break
+
+            if not has_subheader and table.row_count >= 2 and table.data:
+                row_height_estimate = (table_bottom - table_top) / table.row_count
+                header_bottom_estimate = table_top + row_height_estimate
+                second_row_top_estimate = table_top + row_height_estimate * 2
+
+                subheader_lines = []
+                for line in text_lines:
+                    if header_bottom_estimate - 5 <= line['y0'] <= second_row_top_estimate - 5:
+                        if line['x0'] >= table_left - 10 and line['x1'] <= table_right + 10:
+                            # Check (A), (B) pattern
+                            if '(A)' in line['text'] or '(B)' in line['text']:
+                                subheader_lines.append(line)
+
+                if subheader_lines:
+                    table = add_annotation_to_table(table, subheader_lines, 'subheader')
+                    logger.debug(f"[PDF] Added subheader to table on page {page_num + 1}")
+
+            result.append(table)
+
+    result.sort(key=lambda t: (t.page_num, t.bbox[1]))
+    return result
+
+
+def add_annotation_to_table(
+    table: TableInfo,
+    text_lines: List[Dict],
+    position: str
+) -> TableInfo:
+    """
+    Adds annotation rows to a table.
+
+    Args:
+        table: TableInfo object
+        text_lines: List of text line dictionaries
+        position: 'footer' or 'subheader'
+
+    Returns:
+        Updated TableInfo
+    """
+    if not text_lines:
+        return table
+
+    text_lines_sorted = sorted(text_lines, key=lambda l: l['x0'])
+
+    table_width = table.bbox[2] - table.bbox[0]
+    col_width = table_width / table.col_count if table.col_count > 0 else table_width
+
+    new_row = [''] * table.col_count
+
+    for line in text_lines_sorted:
+        relative_x = line['x0'] - table.bbox[0]
+        col_idx = min(int(relative_x / col_width), table.col_count - 1)
+        col_idx = max(0, col_idx)
+
+        if new_row[col_idx]:
+            new_row[col_idx] += " " + line['text']
+        else:
+            new_row[col_idx] = line['text']
+
+    non_empty_cols = sum(1 for c in new_row if c)
+    if non_empty_cols == 1 and new_row[0]:
+        combined_text = " ".join(line['text'] for line in text_lines_sorted)
+        new_row = [combined_text] + [''] * (table.col_count - 1)
+
+    new_data = list(table.data)
+
+    # Update cell info
+    new_cells_info = None
+    if table.cells_info:
+        new_cells_info = list(table.cells_info)
+    else:
+        new_cells_info = []
+
+    if position == 'subheader':
+        if len(new_data) > 0:
+            new_data.insert(1, new_row)
+            # Adjust existing cell info row indices (+1 for row >= 1)
+            adjusted_cells = []
+            for cell in new_cells_info:
+                if cell['row'] >= 1:
+                    adjusted_cell = dict(cell)
+                    adjusted_cell['row'] = cell['row'] + 1
+                    adjusted_cells.append(adjusted_cell)
+                else:
+                    adjusted_cells.append(cell)
+            new_cells_info = adjusted_cells
+            # Add cell info for new subheader row (each cell has colspan=1)
+            for col_idx in range(table.col_count):
+                new_cells_info.append({
+                    'row': 1,
+                    'col': col_idx,
+                    'rowspan': 1,
+                    'colspan': 1,
+                    'bbox': None
+                })
+        else:
+            new_data.append(new_row)
+    else:
+        new_data.append(new_row)
+        # Footer row cell info is handled in generate_html_from_cells
+
+    all_y = [line['y0'] for line in text_lines] + [line['y1'] for line in text_lines]
+    min_y = min(all_y)
+    max_y = max(all_y)
+
+    new_bbox = (
+        table.bbox[0],
+        min(table.bbox[1], min_y),
+        table.bbox[2],
+        max(table.bbox[3], max_y)
+    )
+
+    return TableInfo(
+        page_num=table.page_num,
+        table_idx=table.table_idx,
+        bbox=new_bbox,
+        data=new_data,
+        col_count=table.col_count,
+        row_count=len(new_data),
+        page_height=table.page_height,
+        cells_info=new_cells_info if new_cells_info else None,
+        annotations=table.annotations,
+        detection_strategy=table.detection_strategy,
+        confidence=table.confidence
+    )
+
+
+# ============================================================================
+# Table Merging
+# ============================================================================
+
+def merge_adjacent_tables(tables: List[TableInfo]) -> List[TableInfo]:
+    """
+    Merge adjacent tables.
+
+    Args:
+        tables: List of TableInfo
+
+    Returns:
+        Merged list of TableInfo
+    """
+    if not tables:
+        return tables
+
+    tables_by_page: Dict[int, List[TableInfo]] = defaultdict(list)
+    for table in tables:
+        tables_by_page[table.page_num].append(table)
+
+    merged_result = []
+
+    for page_num, page_tables in tables_by_page.items():
+        sorted_tables = sorted(page_tables, key=lambda t: t.bbox[1])
+
+        i = 0
+        while i < len(sorted_tables):
+            current = sorted_tables[i]
+
+            merged = current
+            while i + 1 < len(sorted_tables):
+                next_table = sorted_tables[i + 1]
+
+                if should_merge_tables(merged, next_table):
+                    merged = do_merge_tables(merged, next_table)
+                    i += 1
+                    logger.debug(f"[PDF] Merged adjacent tables on page {page_num + 1}")
+                else:
+                    break
+
+            merged_result.append(merged)
+            i += 1
+
+    merged_result.sort(key=lambda t: (t.page_num, t.bbox[1]))
+    return merged_result
+
+
+def should_merge_tables(t1: TableInfo, t2: TableInfo) -> bool:
+    """
+    Determine whether two tables should be merged.
+
+    Args:
+        t1: First table
+        t2: Second table
+
+    Returns:
+        True if should merge, False otherwise
+    """
+    if t1.page_num != t2.page_num:
+        return False
+
+    y_gap = t2.bbox[1] - t1.bbox[3]
+    if y_gap < 0 or y_gap > 30:
+        return False
+
+    x_overlap_start = max(t1.bbox[0], t2.bbox[0])
+    x_overlap_end = min(t1.bbox[2], t2.bbox[2])
+    x_overlap = max(0, x_overlap_end - x_overlap_start)
+
+    t1_width = t1.bbox[2] - t1.bbox[0]
+    t2_width = t2.bbox[2] - t2.bbox[0]
+
+    overlap_ratio = x_overlap / max(t1_width, t2_width, 1)
+    if overlap_ratio < 0.8:
+        return False
+
+    if t1.col_count == t2.col_count:
+        return True
+    if t1.row_count == 1 and t1.col_count < t2.col_count:
+        return True
+
+    return False
+
+
+def do_merge_tables(t1: TableInfo, t2: TableInfo) -> TableInfo:
+    """
+    Perform table merging.
+
+    Improvements:
+    - Maintain basic cell info even without cells_info
+    - Accurately adjust cell indices after merging
+
+    Args:
+        t1: First table
+        t2: Second table
+
+    Returns:
+        Merged TableInfo
+    """
+    merged_bbox = (
+        min(t1.bbox[0], t2.bbox[0]),
+        t1.bbox[1],
+        max(t1.bbox[2], t2.bbox[2]),
+        t2.bbox[3]
+    )
+
+    merged_col_count = max(t1.col_count, t2.col_count)
+
+    merged_data = []
+    merged_cells = []
+
+    # Process t1 data
+    t1_row_count = len(t1.data)
+
+    if t1.col_count < merged_col_count and t1.row_count == 1 and t1.data:
+        # Handle colspan when header row has fewer columns
+        extra_cols = merged_col_count - t1.col_count
+        header_row = list(t1.data[0])
+
+        new_header = []
+        col_position = 0
+
+        for orig_col_idx, value in enumerate(header_row):
+            new_header.append(value)
+
+            if orig_col_idx == 1 and extra_cols > 0:
+                colspan = 1 + extra_cols
+                merged_cells.append({
+                    'row': 0,
+                    'col': col_position,
+                    'rowspan': 1,
+                    'colspan': colspan,
+                    'bbox': None
+                })
+                for _ in range(extra_cols):
+                    new_header.append('')
+                col_position += colspan
+            else:
+                merged_cells.append({
+                    'row': 0,
+                    'col': col_position,
+                    'rowspan': 1,
+                    'colspan': 1,
+                    'bbox': None
+                })
+                col_position += 1
+
+        merged_data.append(new_header)
+    else:
+        # Process regular rows
+        for row_idx, row in enumerate(t1.data):
+            if len(row) < merged_col_count:
+                adjusted_row = list(row) + [''] * (merged_col_count - len(row))
+            else:
+                adjusted_row = list(row)
+            merged_data.append(adjusted_row)
+
+        # Copy t1 cell info
+        if t1.cells_info:
+            merged_cells.extend(t1.cells_info)
+
+    # Process t2 data
+    row_offset = t1_row_count
+
+    for row in t2.data:
+        if len(row) < merged_col_count:
+            adjusted_row = list(row) + [''] * (merged_col_count - len(row))
+        else:
+            adjusted_row = list(row)
+        merged_data.append(adjusted_row)
+
+    # Copy t2 cell info (with row offset applied)
+    if t2.cells_info:
+        for cell in t2.cells_info:
+            adjusted_cell = dict(cell)
+            adjusted_cell['row'] = cell.get('row', 0) + row_offset
+            merged_cells.append(adjusted_cell)
+
+    # If cell info is empty, set to None (handled by CellAnalysisEngine)
+    final_cells_info = merged_cells if merged_cells else None
+
+    return TableInfo(
+        page_num=t1.page_num,
+        table_idx=t1.table_idx,
+        bbox=merged_bbox,
+        data=merged_data,
+        col_count=merged_col_count,
+        row_count=len(merged_data),
+        page_height=t1.page_height,
+        cells_info=final_cells_info,
+        detection_strategy=t1.detection_strategy,
+        confidence=max(t1.confidence, t2.confidence)
+    )
+
+
+# ============================================================================
+# Table Continuity Processing
+# ============================================================================
+
+def process_table_continuity(all_tables: List[TableInfo]) -> List[TableInfo]:
+    """
+    Handle table continuity across pages.
+
+    Args:
+        all_tables: List of all TableInfo
+
+    Returns:
+        Processed list of TableInfo
+    """
+    if not all_tables:
+        return all_tables
+
+    result = []
+    last_category = None
+
+    for i, table_info in enumerate(all_tables):
+        table_info = TableInfo(
+            page_num=table_info.page_num,
+            table_idx=table_info.table_idx,
+            bbox=table_info.bbox,
+            data=copy.deepcopy(table_info.data),
+            col_count=table_info.col_count,
+            row_count=table_info.row_count,
+            page_height=table_info.page_height,
+            cells_info=table_info.cells_info,
+            annotations=table_info.annotations,
+            detection_strategy=table_info.detection_strategy,
+            confidence=table_info.confidence
+        )
+
+        curr_data = table_info.data
+
+        if i == 0:
+            last_category = extract_last_category(curr_data)
+            result.append(table_info)
+            continue
+
+        prev_table = all_tables[i - 1]
+
+        is_continuation = (
+            table_info.page_num > prev_table.page_num and
+            prev_table.bbox[3] > prev_table.page_height * 0.7 and
+            table_info.bbox[1] < table_info.page_height * 0.3 and
+            table_info.col_count == prev_table.col_count
+        )
+
+        if is_continuation and last_category:
+            for row in curr_data:
+                if len(row) >= 2:
+                    first_col = row[0]
+                    second_col = row[1] if len(row) > 1 else ""
+
+                    if (not first_col or not str(first_col).strip()) and second_col and str(second_col).strip():
+                        row[0] = last_category
+                    elif first_col and str(first_col).strip():
+                        last_category = first_col
+        else:
+            new_last = extract_last_category(curr_data)
+            if new_last:
+                last_category = new_last
+
+        result.append(table_info)
+
+    return result
+
+
+def extract_last_category(table_data: List[List[Optional[str]]]) -> Optional[str]:
+    """
+    Extract last category from table.
+
+    Args:
+        table_data: Table data
+
+    Returns:
+        Last category string or None
+    """
+    if not table_data:
+        return None
+
+    last_category = None
+
+    for row in table_data:
+        if len(row) >= 1 and row[0] and str(row[0]).strip():
+            last_category = str(row[0]).strip()
+
+    return last_category
+
+
+# ============================================================================
+# HTML Conversion
+# ============================================================================
+
+def is_single_column_table(table_info: TableInfo) -> bool:
+    """
+    Determines if a table has n rows × 1 column format.
+
+    Tables with n rows × 1 column are often not actual tables,
+    so converting them to a text list is more appropriate.
+
+    Args:
+        table_info: Table information
+
+    Returns:
+        True if single-column table, False otherwise
+    """
+    data = table_info.data
+
+    if not data:
+        return False
+
+    # Calculate max columns across all rows
+    max_cols = max(len(row) for row in data) if data else 0
+
+    # Single column if max_cols is 1
+    return max_cols == 1
+
+
+def convert_single_column_to_text(table_info: TableInfo) -> str:
+    """
+    Converts a single-column table to a text list.
+
+    Data with n rows × 1 column format is semantically more
+    appropriate to express as structured text rather than a table.
+
+    Args:
+        table_info: Table information
+
+    Returns:
+        String in text list format
+    """
+    data = table_info.data
+
+    if not data:
+        return ""
+
+    lines = []
+    for row in data:
+        if row and len(row) > 0:
+            cell_text = str(row[0]).strip() if row[0] else ""
+            if cell_text:
+                lines.append(cell_text)
+
+    return '\n'.join(lines)
+
+
+def convert_table_to_html(table_info: TableInfo) -> str:
+    """
+    Converts a table to HTML.
+
+    Improvements:
+    1. Prioritize using PyMuPDF cell info
+    2. Apply CellAnalysisEngine
+    3. Accurate rowspan/colspan handling
+    4. Full colspan for annotation rows
+    5. Semantic HTML with accessibility considerations
+
+    Args:
+        table_info: Table information
+
+    Returns:
+        HTML string
+    """
+    data = table_info.data
+
+    if not data:
+        return ""
+
+    num_rows = len(data)
+    num_cols = max(len(row) for row in data) if data else 0
+
+    if num_cols == 0:
+        return ""
+
+    # Perform cell analysis using CellAnalysisEngine
+    cell_engine = CellAnalysisEngine(table_info, None)
+    analyzed_cells = cell_engine.analyze()
+
+    # Generate HTML from analyzed cell info
+    return generate_html_from_cells(data, analyzed_cells, num_rows, num_cols)
+
+
+def generate_html_from_cells(
+    data: List[List[Optional[str]]],
+    cells_info: List[Dict],
+    num_rows: int,
+    num_cols: int
+) -> str:
+    """
+    Improved HTML generation.
+
+    Improvements:
+    - Process all cells even with incomplete cell info
+    - Render empty cells correctly
+    - Enhanced data range validation
+
+    Args:
+        data: Table data
+        cells_info: Cell information list
+        num_rows: Number of rows
+        num_cols: Number of columns
+
+    Returns:
+        HTML string
+    """
+    # Create span_map: (row, col) -> {rowspan, colspan}
+    span_map: Dict[Tuple[int, int], Dict] = {}
+
+    for cell in cells_info:
+        row = cell.get('row', 0)
+        col = cell.get('col', 0)
+        rowspan = max(1, cell.get('rowspan', 1))
+        colspan = max(1, cell.get('colspan', 1))
+
+        # Adjust to stay within data range
+        if row >= num_rows or col >= num_cols:
+            continue
+
+        rowspan = min(rowspan, num_rows - row)
+        colspan = min(colspan, num_cols - col)
+
+        key = (row, col)
+        span_map[key] = {
+            'rowspan': rowspan,
+            'colspan': colspan
+        }
+
+    # Create skip_set: positions covered by merged cells
+    skip_set: Set[Tuple[int, int]] = set()
+
+    for (row, col), spans in span_map.items():
+        rowspan = spans['rowspan']
+        colspan = spans['colspan']
+
+        for r in range(row, min(row + rowspan, num_rows)):
+            for c in range(col, min(col + colspan, num_cols)):
+                if (r, c) != (row, col):
+                    skip_set.add((r, c))
+
+    # Detect annotation rows and apply full colspan
+    for row_idx, row in enumerate(data):
+        if not row:
+            continue
+        first_val = str(row[0]).strip() if row[0] else ""
+
+        is_annotation = False
+        for pattern in PDFConfig.ANNOTATION_PATTERNS:
+            if first_val.startswith(pattern):
+                is_annotation = True
+                break
+
+        if is_annotation:
+            # Annotation row gets full colspan
+            span_map[(row_idx, 0)] = {'rowspan': 1, 'colspan': num_cols}
+            for col_idx in range(1, num_cols):
+                skip_set.add((row_idx, col_idx))
+
+    # Generate HTML
+    html_parts = ["<table>"]
+
+    for row_idx in range(num_rows):
+        html_parts.append("  <tr>")
+
+        row_data = data[row_idx] if row_idx < len(data) else []
+
+        for col_idx in range(num_cols):
+            # Check if this cell should be skipped
+            if (row_idx, col_idx) in skip_set:
+                continue
+
+            # Extract cell content
+            content = ""
+            if col_idx < len(row_data):
+                content = row_data[col_idx]
+            content = escape_html(str(content).strip() if content else "")
+
+            # Get span info (default to 1 if not found)
+            spans = span_map.get((row_idx, col_idx), {'rowspan': 1, 'colspan': 1})
+            attrs = []
+
+            if spans['rowspan'] > 1:
+                attrs.append(f'rowspan="{spans["rowspan"]}"')
+            if spans['colspan'] > 1:
+                attrs.append(f'colspan="{spans["colspan"]}"')
+
+            attr_str = " " + " ".join(attrs) if attrs else ""
+
+            # First row is treated as header
+            tag = "th" if row_idx == 0 else "td"
+            html_parts.append(f"    <{tag}{attr_str}>{content}</{tag}>")
+
+        html_parts.append("  </tr>")
+
+    html_parts.append("</table>")
+    return "\n".join(html_parts)
+