MinerU/zhch/文字PDF-表格处理.md

正在收集工作区信息正在筛选到最相关的信息根据代码分析,表格处理的代码主要在以下几个文件中:

表格处理流程代码位置

1. 表格识别入口 - batch_analyze.py

# 1. 裁剪表格区域图像
def get_crop_table_img(scale):
    crop_xmin, crop_ymin = int(table_res['poly'][0]), int(table_res['poly'][1])
    crop_xmax, crop_ymax = int(table_res['poly'][4]), int(table_res['poly'][5])
    bbox = (int(crop_xmin / scale), int(crop_ymin / scale), 
            int(crop_xmax / scale), int(crop_ymax / scale))
    return get_crop_np_img(bbox, np_img, scale=scale)

wireless_table_img = get_crop_table_img(scale = 1)
wired_table_img = get_crop_table_img(scale = 10/3)

# 2. 表格分类（有线/无线表格）
table_cls_model = atom_model_manager.get_atom_model(
    atom_model_name=AtomicModel.TableCls,
)
table_cls_model.batch_predict(table_res_list_all_page, batch_size=16)

# 3. OCR检测 - 获取文本框
det_ocr_engine = atom_model_manager.get_atom_model(
    atom_model_name=AtomicModel.OCR,
    det_db_box_thresh=0.5,
    det_db_unclip_ratio=1.6,
    enable_merge_det_boxes=False,
)
for table_res_dict in table_res_list_all_page:
    bgr_image = cv2.cvtColor(table_res_dict["table_img"], cv2.COLOR_RGB2BGR)
    ocr_result = det_ocr_engine.ocr(bgr_image, rec=False)[0]  # 只检测不识别
    # 构造dt_box列表

2. OCR识别 - batch_analyze.py (line 157-178)

# 4. OCR识别 - 识别文本内容
for _lang, rec_img_list in rec_img_lang_group.items():
    ocr_engine = atom_model_manager.get_atom_model(
        atom_model_name=AtomicModel.OCR,
        det_db_box_thresh=0.5,
        det_db_unclip_ratio=1.6,
        lang=_lang,
        enable_merge_det_boxes=False,
    )
    cropped_img_list = [item["cropped_img"] for item in rec_img_list]
    ocr_res_list = ocr_engine.ocr(cropped_img_list, det=False, tqdm_enable=True)[0]
    
    # 回填OCR结果
    for img_dict, ocr_res in zip(rec_img_list, ocr_res_list):
        table_res_list_all_page[img_dict["table_id"]]["ocr_result"].append(
            [img_dict["dt_box"], html.escape(ocr_res[0]), ocr_res[1]]
        )

3. 无线表格识别 - batch_analyze.py (line 192-195)

# 5. 无线表格模型预测（所有表格先用无线表格模型）
wireless_table_model = atom_model_manager.get_atom_model(
    atom_model_name=AtomicModel.WirelessTable,
)
wireless_table_model.batch_predict(table_res_list_all_page)

对应的模型代码在 RapidTableModel:

def batch_predict(self, table_res_list: List[Dict], batch_size: int = 4) -> None:
    """对传入的字典列表进行批量预测，无返回值"""
    
    for index in range(0, len(not_none_table_res_list), batch_size):
        batch_imgs = [cv2.cvtColor(np.asarray(table[i]["table_img"]), 
                                   cv2.COLOR_RGB2BGR) 
                     for i in range(index, min(index + batch_size, len(...)))]
        batch_ocrs = [table[i]["ocr_result"] 
                     for i in range(index, min(index + batch_size, len(...)))]
        
        # 调用RapidTable模型
        batch_results = self.table_model.batch_predict(batch_imgs, batch_ocrs, batch_size)
        
        # 更新结果
        for i, result in enumerate(batch_results):
            not_none_table_res_list[index + i]['table_res']['html'] = result.pred_html

4. 有线表格识别 - batch_analyze.py (line 204-225)

# 6. 有线表格模型预测（针对分类为有线的表格）
if wired_table_res_list:
    for table_res_dict in tqdm(wired_table_res_list, desc="Table-wired Predict"):
        wired_table_model = atom_model_manager.get_atom_model(
            atom_model_name=AtomicModel.WiredTable,
            lang=table_res_dict["lang"],
        )
        # 传入无线表格的HTML作为参考
        table_res_dict["table_res"]["html"] = wired_table_model.predict(
            table_res_dict["wired_table_img"],
            table_res_dict["ocr_result"],
            table_res_dict["table_res"].get("html", None)  # 无线表格HTML
        )

对应的模型代码在 UnetTableModel:

def predict(self, input_img, ocr_result, wireless_html_code):
    # 7. 生成有线表格HTML
    wired_table_results = self.wired_table_model(np_img, ocr_result)
    wired_html_code = wired_table_results.pred_html
    
    # 8. 比较有线/无线表格结果，选择更好的
    wired_len = count_table_cells_physical(wired_html_code)
    wireless_len = count_table_cells_physical(wireless_html_code)
    
    # 计算非空单元格数量
    wireless_non_blank_count = wireless_len - wireless_blank_count
    wired_non_blank_count = wired_len - wired_blank_count
    
    # 选择更优结果
    if wireless_non_blank_count > wired_non_blank_count:
        return wireless_html_code
    else:
        return wired_html_code

5. 表格HTML生成 - WiredTableRecognition

def __call__(self, img, ocr_result=None, **kwargs):
    # 1. 表格结构检测
    polygons, rotated_polygons = self.table_structure(img, **kwargs)
    
    # 2. 表格恢复（生成逻辑点和单元格映射）
    table_res, logi_points = self.table_recover(
        rotated_polygons, row_threshold, col_threshold
    )
    
    # 3. OCR结果匹配到单元格
    cell_box_det_map = match_ocr_cell(ocr_result, rotated_polygons, logi_points)
    
    # 4. 补充空单元格的OCR识别
    cell_box_det_map = self.fill_blank_rec(img, polygons, cell_box_det_map)
    
    # 5. 排序和合并OCR结果
    t_rec_ocr_list = self.sort_and_gather_ocr_res(t_rec_ocr_list)
    
    # 6. 生成HTML
    pred_html = plot_html_table(logi_points, cell_box_det_map)
    
    return WiredTableOutput(pred_html, polygons, logi_points, elapse)

HTML生成的具体实现在 plot_html_table:

def plot_html_table(logi_points, cell_box_map) -> str:
    table_html = "<html><body><table>"
    
    # 遍历每行
    for row in range(max_row):
        temp = "<tr>"
        for col in range(max_col):
            if grid[row][col]:
                i, row_start, row_end, col_start, col_end = grid[row][col]
                if row == row_start and col == col_start:
                    # 获取单元格文本
                    text = "".join(cell_box_map.get(i))
                    
                    # 计算跨行跨列
                    row_span = row_end - row_start + 1
                    col_span = col_end - col_start + 1
                    
                    # 生成HTML单元格
                    cell_content = f"<td rowspan={row_span} colspan={col_span}>{text}</td>"
                    temp += cell_content
        table_html = table_html + temp + "</tr>"
    
    table_html += "</table></body></html>"
    return table_html

完整流程总结

graph LR
    A[裁剪表格图像] --> B[表格分类<br/>有线/无线]
    B --> C[OCR检测文本框]
    C --> D[OCR识别文本]
    D --> E[无线表格识别<br/>RapidTable]
    E --> F{是否有线表格?}
    F -->|是| G[有线表格识别<br/>UnetTable]
    F -->|否| H[直接使用无线结果]
    G --> I[比较有线/无线结果]
    I --> J[选择更优HTML]
    H --> J
    J --> K[清理HTML格式]
    K --> L[返回table HTML]

关键代码文件：

• 表格检测和分类: batch_analyze.py
• 无线表格模型: slanet_plus/main.py
• 有线表格模型: unet_table/main.py
• HTML生成: utils_table_recover.py