docvortex 0.2.1__py3-none-any.whl

docvortex/analyzers/native/pdf/table_text_styles.py

@@ -0,0 +1,283 @@
+"""使用 PDF 原生字符几何为高置信表格 HTML 恢复上下标。"""
+
+from __future__ import annotations
+
+import html
+from collections import defaultdict
+from typing import Any
+
+from ....document.pdf.text.contracts import Char
+
+from ....schema import BBox
+from .geometry import _bbox_union_many, _coerce_bbox
+from .line_merging import _merge_overlapping_inline_text_clusters
+from .models import _LineItem
+from .native_text import _fill_native_typography
+from .script_geometry import ScriptRole
+from .table_recovery.candidate import serialize_native_table_html
+from .table_recovery.contracts import (
+    NativeTableCell,
+    NativeTableGlyph,
+    NativeTableInput,
+    NativeTableResult,
+    NativeTableRule,
+    NativeTableText,
+)
+from .table_recovery.geometry import page_bbox_to_table_local
+from .table_recovery.text import build_cell_text_parts
+from .inline.scripts import _fraction_member_indices, _script_line_char_roles
+
+
+def _table_char_map(chars: tuple[Char, ...]) -> dict[int, Char]:
+    """按合法 char_idx 建立页面字符查询表，重复索引保留首项。"""
+
+    output: dict[int, Char] = {}
+    for fallback_index, char in enumerate(chars):
+        char_idx = char.get("char_idx", fallback_index)
+        if isinstance(char_idx, bool) or not isinstance(char_idx, int):
+            continue
+        output.setdefault(char_idx, char)
+    return output
+
+
+def _cell_glyphs(
+    result: NativeTableResult,
+    cell: NativeTableCell,
+    glyph_by_source: dict[int, NativeTableGlyph] | None = None,
+) -> list[NativeTableGlyph]:
+    """按 cell 的稳定字符来源收集并恢复视觉行内顺序。"""
+
+    if glyph_by_source is None:
+        glyph_by_source = {glyph.source_index: glyph for glyph in result.text.glyphs}
+    return sorted(
+        (glyph_by_source[source_index] for source_index in cell.source_char_indices if source_index in glyph_by_source),
+        key=lambda glyph: (glyph.visual_row, glyph.bbox[0], glyph.bbox[1], glyph.glyph_id),
+    )
+
+
+def _cell_visual_lines(
+    glyphs: list[NativeTableGlyph],
+    chars_by_source: dict[int, Char],
+    page_size: tuple[float, float],
+    angle: int,
+) -> list[_LineItem]:
+    """把一个 cell 的 glyph 按 visual row 转换成正文公式分段使用的行。"""
+
+    grouped: dict[int, list[NativeTableGlyph]] = defaultdict(list)
+    for glyph in glyphs:
+        grouped[glyph.visual_row].append(glyph)
+
+    lines: list[_LineItem] = []
+    for visual_row, row_glyphs in sorted(grouped.items()):
+        ordered = sorted(row_glyphs, key=lambda glyph: (glyph.bbox[0], glyph.bbox[1], glyph.glyph_id))
+        chars = [chars_by_source[glyph.source_index] for glyph in ordered if glyph.source_index in chars_by_source]
+        bboxes = [bbox for char in chars if (bbox := _coerce_bbox(char.get("bbox"))) is not None]
+        if not chars or not bboxes:
+            continue
+        line = _LineItem(
+            text="".join(glyph.text for glyph in ordered),
+            bbox=_bbox_union_many(bboxes),
+            angle=angle,
+            source_index=visual_row,
+            chars=chars,
+            visual_row_id=visual_row,
+        )
+        _fill_native_typography(line, page_size)
+        lines.append(line)
+    return lines
+
+
+def _rule_overlaps_boundary(
+    rule_bbox: BBox,
+    boundary_y: float,
+    boundary_left: float,
+    boundary_right: float,
+    tolerance: float,
+) -> bool:
+    """判断局部横线是否与一个逻辑 cell 的水平边界重合。"""
+
+    rule_y = (rule_bbox[1] + rule_bbox[3]) / 2.0
+    if abs(rule_y - boundary_y) > tolerance:
+        return False
+    overlap = min(rule_bbox[2], boundary_right) - max(rule_bbox[0], boundary_left)
+    rule_width = rule_bbox[2] - rule_bbox[0]
+    boundary_width = boundary_right - boundary_left
+    return overlap > 0 and overlap / max(1e-6, min(rule_width, boundary_width)) >= 0.5
+
+
+def _non_grid_fraction_rules(
+    table_input: NativeTableInput,
+    result: NativeTableResult,
+) -> list[NativeTableRule]:
+    """移除与恢复后 cell 边界重合的横线，仅保留可能的分式线。"""
+
+    tolerance = max(0.75, 0.12 * result.text.median_glyph_height)
+    output: list[NativeTableRule] = []
+    for rule in table_input.drawing_lines:
+        if rule.orientation != "horizontal":
+            continue
+        local_bbox = page_bbox_to_table_local(rule.bbox, table_input.table_bbox, table_input.angle)
+        if local_bbox is None:
+            continue
+        is_grid_rule = any(
+            _rule_overlaps_boundary(
+                local_bbox,
+                boundary_y,
+                cell.bbox[0],
+                cell.bbox[2],
+                max(tolerance, rule.width),
+            )
+            for cell in result.cells
+            for boundary_y in (cell.bbox[1], cell.bbox[3])
+        )
+        if not is_grid_rule:
+            output.append(rule)
+    return output
+
+
+def _drop_shifted_word_prefixes(chars: list[dict[str, Any]], roles: list[ScriptRole]) -> None:
+    """拒绝普通字母词仅首字母偏移的弱候选，完整同基线词由精炼层闭合。"""
+
+    start = 0
+    while start < len(chars):
+        if not str(chars[start].get("char", "")).isalpha():
+            start += 1
+            continue
+        end = start + 1
+        while end < len(chars) and str(chars[end].get("char", "")).isalpha():
+            end += 1
+        if end - start >= 2 and roles[start] != "body" and all(role == "body" for role in roles[start + 1 : end]):
+            roles[start] = "body"
+        start = end
+
+
+def _cell_script_roles(
+    glyphs: list[NativeTableGlyph],
+    chars_by_source: dict[int, Char],
+    page_size: tuple[float, float],
+    angle: int,
+    tight_bboxes: dict[int, BBox],
+    origins: dict[int, tuple[float, float]],
+    fraction_members: set[int],
+) -> dict[int, ScriptRole]:
+    """在 cell 内按正文同款二维公式分段和字符几何返回稳定角色。"""
+
+    lines = _cell_visual_lines(glyphs, chars_by_source, page_size, angle)
+    if not lines:
+        return {}
+    segmented_lines = _merge_overlapping_inline_text_clusters(lines, page_size, [])
+    roles_by_source: dict[int, ScriptRole] = {}
+    for line in segmented_lines:
+        chars, roles, _body_counts, _formula_flags = _script_line_char_roles(
+            line,
+            page_size,
+            tight_bboxes,
+            origins,
+            fraction_members,
+        )
+        _drop_shifted_word_prefixes(chars, roles)
+        for char, role in zip(chars, roles, strict=True):
+            char_idx = char.get("char_idx")
+            if not isinstance(char_idx, int) or role == "body":
+                continue
+            previous = roles_by_source.get(char_idx)
+            roles_by_source[char_idx] = role if previous in {None, role} else "body"
+    return {source_index: role for source_index, role in roles_by_source.items() if role != "body"}
+
+
+def _render_styled_cell(
+    cell: NativeTableCell,
+    glyphs: list[NativeTableGlyph],
+    median_height: float,
+    roles: dict[int, ScriptRole],
+) -> str:
+    """按旧文本重建规则安全插入平坦的 sup/sub 标签。"""
+
+    parts = build_cell_text_parts(glyphs, median_height)
+    if "".join(text for text, _source_index in parts) != cell.content:
+        return html.escape(cell.content, quote=False)
+
+    rendered: list[str] = []
+    active_role: ScriptRole = "body"
+    active_parts: list[str] = []
+
+    def flush() -> None:
+        """提交当前同角色片段，并只生成受信的上下标标签。"""
+
+        nonlocal active_parts
+        if not active_parts:
+            return
+        content = html.escape("".join(active_parts), quote=False)
+        if active_role == "sup":
+            rendered.append(f"<sup>{content}</sup>")
+        elif active_role == "sub":
+            rendered.append(f"<sub>{content}</sub>")
+        else:
+            rendered.append(content)
+        active_parts = []
+
+    for text, source_index in parts:
+        role = roles.get(source_index, "body") if source_index is not None else "body"
+        if role != active_role:
+            flush()
+            active_role = role
+        active_parts.append(text)
+    flush()
+    return "".join(rendered)
+
+
+def render_native_table_html_with_scripts(
+    result: NativeTableResult,
+    table_input: NativeTableInput,
+    tight_bboxes: dict[int, BBox],
+    origins: dict[int, tuple[float, float]],
+) -> str:
+    """为高置信原生表格恢复上下标，证据不足时返回原始 HTML。"""
+
+    if not tight_bboxes or not origins:
+        return result.html
+    if not isinstance(getattr(result, "text", None), NativeTableText):
+        return result.html
+    chars_by_source = _table_char_map(table_input.chars)
+    fraction_rules = _non_grid_fraction_rules(table_input, result)
+    # 一张表只建立一次来源索引；保持原字典推导式的后项覆盖语义。
+    glyph_by_source = {glyph.source_index: glyph for glyph in result.text.glyphs}
+    cell_glyphs = {(cell.row, cell.col): _cell_glyphs(result, cell, glyph_by_source) for cell in result.cells}
+    cell_roles: dict[tuple[int, int], dict[int, ScriptRole]] = {}
+    for cell in result.cells:
+        key = (cell.row, cell.col)
+        glyphs = cell_glyphs[key]
+        cell_chars = [chars_by_source[glyph.source_index] for glyph in glyphs if glyph.source_index in chars_by_source]
+        fraction_members = _fraction_member_indices(
+            table_input.page_size,
+            cell_chars,
+            tight_bboxes,
+            fraction_rules,
+            table_input.angle,
+        )
+        roles = _cell_script_roles(
+            glyphs,
+            chars_by_source,
+            table_input.page_size,
+            table_input.angle,
+            tight_bboxes,
+            origins,
+            fraction_members,
+        )
+        if roles:
+            cell_roles[key] = roles
+    if not cell_roles:
+        return result.html
+    return serialize_native_table_html(
+        result.rows,
+        result.cells,
+        render_cell=lambda cell: _render_styled_cell(
+            cell,
+            cell_glyphs[(cell.row, cell.col)],
+            result.text.median_glyph_height,
+            cell_roles.get((cell.row, cell.col), {}),
+        ),
+    )
+
+
+__all__ = ["render_native_table_html_with_scripts"]

docvortex/analyzers/native/pdf/tables.py

@@ -0,0 +1,147 @@
+"""PDF 表格的兼容入口；实现按检测、注释和物化职责组织。"""
+
+from .table_annotations import (
+    _build_table_annotation as _build_table_annotation,
+    _collect_caption_rows as _collect_caption_rows,
+    _collect_footnote_rows as _collect_footnote_rows,
+    _extract_auxiliary_table_note_marker as _extract_auxiliary_table_note_marker,
+    _table_core_references_marker as _table_core_references_marker,
+    _line_has_compact_marker_token as _line_has_compact_marker_token,
+    _line_has_superscript_marker as _line_has_superscript_marker,
+    _table_note_body_reference_height as _table_note_body_reference_height,
+    _visual_row_text as _visual_row_text,
+    _is_table_note_text as _is_table_note_text,
+    _find_table_caption as _find_table_caption,
+    _find_caption_number_peers as _find_caption_number_peers,
+    _merge_table_candidate_annotations as _merge_table_candidate_annotations,
+)
+from .table_constants import (
+    _TABLE_CAPTION_RE as _TABLE_CAPTION_RE,
+    _TABLE_CONTINUATION_RE as _TABLE_CONTINUATION_RE,
+    _TABLE_NOTE_RE as _TABLE_NOTE_RE,
+    _AUXILIARY_TABLE_NOTE_RE as _AUXILIARY_TABLE_NOTE_RE,
+    _TABLE_SPLIT_NUMBER_RE as _TABLE_SPLIT_NUMBER_RE,
+    _FILLED_GRID_MIN_PAGE_AREA_RATIO as _FILLED_GRID_MIN_PAGE_AREA_RATIO,
+    _FILLED_GRID_MAX_PAGE_AREA_RATIO as _FILLED_GRID_MAX_PAGE_AREA_RATIO,
+    _FILLED_GRID_MIN_PAGE_WIDTH_RATIO as _FILLED_GRID_MIN_PAGE_WIDTH_RATIO,
+    _FILLED_GRID_MIN_PAGE_HEIGHT_RATIO as _FILLED_GRID_MIN_PAGE_HEIGHT_RATIO,
+)
+from .table_detection import (
+    _detect_table_candidates as _detect_table_candidates,
+    _externalize_table_continuation_captions as _externalize_table_continuation_captions,
+)
+from .table_filled_grid import (
+    _detect_filled_grid_table_candidates as _detect_filled_grid_table_candidates,
+    _select_maximal_filled_grid_cells as _select_maximal_filled_grid_cells,
+    _bbox_is_contained_with_tolerance as _bbox_is_contained_with_tolerance,
+    _group_filled_grid_cells_into_bands as _group_filled_grid_cells_into_bands,
+    _filled_grid_band_covers_outer_width as _filled_grid_band_covers_outer_width,
+)
+from .table_rows import (
+    _clip_visual_row_to_corridor as _clip_visual_row_to_corridor,
+)
+from .table_materialization import (
+    _recover_native_table_html as _recover_native_table_html,
+    _materialize_table_blocks as _materialize_table_blocks,
+    _materialize_table_annotations as _materialize_table_annotations,
+    _split_table_annotation_visual_groups as _split_table_annotation_visual_groups,
+    _build_table_annotation_blocks as _build_table_annotation_blocks,
+    _merge_table_annotation_content as _merge_table_annotation_content,
+    _table_body_materialization_bbox as _table_body_materialization_bbox,
+    _candidate_projection_line_indices as _candidate_projection_line_indices,
+    _expand_candidate_same_baseline_members as _expand_candidate_same_baseline_members,
+)
+from .table_rules import (
+    _build_fragments as _build_fragments,
+    _cluster_fragment_rows as _cluster_fragment_rows,
+    _build_rule_table_candidates as _build_rule_table_candidates,
+    _expand_candidates_to_connected_rule_grids as _expand_candidates_to_connected_rule_grids,
+    _build_closed_rule_grid_candidates as _build_closed_rule_grid_candidates,
+    _closed_grid_vertical_track_positions as _closed_grid_vertical_track_positions,
+    _count_occupied_closed_grid_columns as _count_occupied_closed_grid_columns,
+    _connected_rule_grid_bboxes as _connected_rule_grid_bboxes,
+    _connected_rule_grid_components as _connected_rule_grid_components,
+    _connected_horizontal_rule_bboxes as _connected_horizontal_rule_bboxes,
+    _rule_bands_share_grid_tracks as _rule_bands_share_grid_tracks,
+    _group_long_horizontal_rules as _group_long_horizontal_rules,
+    _rows_inside_rule_interval as _rows_inside_rule_interval,
+    _every_rule_interval_has_multi_cell_row as _every_rule_interval_has_multi_cell_row,
+    _rule_intervals_are_column_compatible as _rule_intervals_are_column_compatible,
+    _continuous_table_row_segments as _continuous_table_row_segments,
+    _table_segment_reaches_boundaries as _table_segment_reaches_boundaries,
+    _table_rows_align_with_rule_span as _table_rows_align_with_rule_span,
+    _count_aligned_vertical_rules as _count_aligned_vertical_rules,
+    _compact_fully_ruled_grid_column_count as _compact_fully_ruled_grid_column_count,
+    _full_height_vertical_rule_positions as _full_height_vertical_rule_positions,
+    _looks_like_page_column_prose as _looks_like_page_column_prose,
+    _count_repeated_fill_bands as _count_repeated_fill_bands,
+    _longest_dense_multi_cell_rows as _longest_dense_multi_cell_rows,
+    _expand_rule_table_candidate as _expand_rule_table_candidate,
+    _count_stable_columns as _count_stable_columns,
+    _merge_table_candidates as _merge_table_candidates,
+    _median_fragment_height as _median_fragment_height,
+)
+from ....document.pdf.text.contracts import (
+    Char as Char,
+)
+from .table_recovery import (
+    NativeTableInput as NativeTableInput,
+    NativeTableRectangle as NativeTableRectangle,
+    NativeTableRule as NativeTableRule,
+    coerce_native_table_rectangles as coerce_native_table_rectangles,
+    coerce_native_table_rules as coerce_native_table_rules,
+    recover_native_pdf_table as recover_native_pdf_table,
+)
+from .table_text_styles import (
+    render_native_table_html_with_scripts as render_native_table_html_with_scripts,
+)
+from ....foundation.text import (
+    merge_text_line_contents as merge_text_line_contents,
+)
+from .spatial_text import (
+    project_pdf_table_text as project_pdf_table_text,
+)
+from ....schema import (
+    BBox as BBox,
+)
+from ....document.pdf.document import (
+    PDFPathInfo as PDFPathInfo,
+)
+from ....foundation.language import (
+    detect_lang as detect_lang,
+)
+from .models import (
+    _Fragment as _Fragment,
+    _LineItem as _LineItem,
+    _LocalAxisLine as _LocalAxisLine,
+    _PageSource as _PageSource,
+    _TableAnnotation as _TableAnnotation,
+    _TableCandidate as _TableCandidate,
+    _VisualRow as _VisualRow,
+)
+from .geometry import (
+    _bbox_area as _bbox_area,
+    _bbox_axis_overlap_ratio as _bbox_axis_overlap_ratio,
+    _bbox_center_x as _bbox_center_x,
+    _bbox_center_y as _bbox_center_y,
+    _bbox_overlap_in_smaller as _bbox_overlap_in_smaller,
+    _bbox_union as _bbox_union,
+    _bbox_union_many as _bbox_union_many,
+    _coerce_bbox as _coerce_bbox,
+    _expand_bbox as _expand_bbox,
+    _point_in_bbox as _point_in_bbox,
+    _rotate_bbox_from_upright as _rotate_bbox_from_upright,
+    _rotate_bbox_to_upright as _rotate_bbox_to_upright,
+    _transform_axis_lines as _transform_axis_lines,
+)
+from .line_layout import (
+    _font_signatures_share_family as _font_signatures_share_family,
+    _line_effective_height as _line_effective_height,
+    _line_tight_output_bbox as _line_tight_output_bbox,
+)
+from .line_merging import (
+    _same_baseline_geometry as _same_baseline_geometry,
+)
+from .native_text import (
+    _normalize_native_run_text as _normalize_native_run_text,
+)

docvortex/analyzers/native/pdf/text_assembly/__init__.py

@@ -0,0 +1,3 @@
+"""按职责拆分的 Flash PDF 内部实现。"""
+
+__all__: list[str] = []