From 6725648080c436487a5c22fbd25cc710e4069ded Mon Sep 17 00:00:00 2001 From: myhloli Date: Mon, 26 Jan 2026 15:21:48 +0800 Subject: [PATCH] feat: enhance DOCX processing to classify caption blocks and clean table HTML --- mineru/backend/office/office_magic_model.py | 161 +++++++++++++++++++- 1 file changed, 156 insertions(+), 5 deletions(-) diff --git a/mineru/backend/office/office_magic_model.py b/mineru/backend/office/office_magic_model.py index 9f1d72a1..1eb24e7a 100644 --- a/mineru/backend/office/office_magic_model.py +++ b/mineru/backend/office/office_magic_model.py @@ -5,7 +5,6 @@ from loguru import logger from mineru.utils.boxbase import calculate_overlap_area_in_bbox1_area_ratio from mineru.utils.enum_class import ContentType, BlockType -from mineru.utils.guess_suffix_or_lang import guess_language_by_text from mineru.utils.magic_model_utils import reduct_overlap, tie_up_category_by_index @@ -15,6 +14,10 @@ class MagicModel: blocks = [] self.all_spans = [] + + # 对caption块进行分类，将其分类为image_caption或table_caption + page_blocks = classify_caption_blocks(page_blocks) + # 解析每个块 for index, block_info in enumerate(page_blocks): @@ -24,7 +27,8 @@ class MagicModel: if block_type in [ "text", "title", - "caption", + "image_caption", + "table_caption", "header", "footer", ]: @@ -44,13 +48,14 @@ class MagicModel: "type": span_type, } if span_type == ContentType.TABLE: - span["html"] = block_info.get("content", "") + span["html"] = clean_table_html(block_info.get("content", "")) elif span_type == ContentType.IMAGE: + # jpg格式base64 span["image_base64"] = block_info.get("content", "") elif span_type in [ContentType.INTERLINE_EQUATION]: span = { "type": span_type, - "content": block_content, + "content": block_info.get("content", ""), } else: @@ -217,6 +222,80 @@ class MagicModel: return self.all_spans +def clean_table_html(html: str) -> str: + """ + 清洗表格HTML，只保留对表格结构表示有用的信息。 + + 保留的属性： + - colspan: 列合并 + - rowspan: 行合并 + + 清洗的内容： + - 移除所有style属性 + - 移除所有class属性 + - 移除border等其他属性 + - 保持表格结构标签（table, thead, tbody, tr, th, td等） + + Args: + html: 原始表格HTML字符串 + + Returns: + 清洗后的HTML字符串 + """ + if not html: + return "" + + # 需要保留的属性（对表格结构有用） + preserved_attrs = {'colspan', 'rowspan'} + + def clean_tag(match): + """清洗单个标签，只保留结构相关的属性""" + full_tag = match.group(0) + tag_name = match.group(1).lower() + + # 自闭合标签的处理 + is_self_closing = full_tag.rstrip().endswith('/>') + + # 提取需要保留的属性 + kept_attrs = [] + + # 匹配所有属性: attr="value" 或 attr='value' 或 attr=value 或单独的attr + attr_pattern = r'(\w+)\s*=\s*(?:"([^"]*)"|\'([^\']*)\'|(\S+))|(\w+)(?=\s|>|/>)' + for attr_match in re.finditer(attr_pattern, full_tag): + if attr_match.group(5): + # 单独的属性（如 disabled），跳过 + continue + + attr_name = attr_match.group(1) + if attr_name is None: + continue + attr_name = attr_name.lower() + attr_value = attr_match.group(2) or attr_match.group(3) or attr_match.group(4) or "" + + # 只保留colspan和rowspan + if attr_name in preserved_attrs: + kept_attrs.append(f'{attr_name}="{attr_value}"') + + # 重建标签 + if kept_attrs: + attrs_str = ' ' + ' '.join(kept_attrs) + else: + attrs_str = '' + + if is_self_closing: + return f'<{tag_name}{attrs_str}/>' + else: + return f'<{tag_name}{attrs_str}>' + + # 匹配开始标签（包括自闭合标签），捕获标签名 + # 匹配 或 + tag_pattern = r'<(\w+)(?:\s+[^>]*)?\s*/?>' + + result = re.sub(tag_pattern, clean_tag, html) + + return result + + def isolated_formula_clean(txt): latex = txt[:] if latex.startswith("\\["): latex = latex[2:] @@ -485,4 +564,76 @@ def fix_list_blocks(list_blocks, text_blocks, ref_text_blocks): else: list_block["sub_type"] = "unknown" - return list_blocks, text_blocks, ref_text_blocks \ No newline at end of file + return list_blocks, text_blocks, ref_text_blocks + + +def classify_caption_blocks(page_blocks: list) -> list: + """ + 对page_blocks中的caption块进行分类，将其分类为image_caption或table_caption。 + + 规则： + 1. 只有与type为table或image相邻的caption可以作为caption + 2. caption块与table或image中相隔的块全部是caption的情况视为该caption块与table或image相邻 + 3. caption的类型与他前置位相邻的母块type一致（table或image），如果没有前置位母块则检查是否有后置位母块 + 4. 没有相邻母块的caption需要变更type为text + """ + if not page_blocks: + return page_blocks + + available_types = ["table", "image"] + + result_blocks = [] + n = len(page_blocks) + + for i, block in enumerate(page_blocks): + if block.get("type") != "caption": + result_blocks.append(block) + continue + + # 查找前置位相邻的母块（table或image） + # 向前查找，跳过连续的caption块 + prev_parent_type = None + j = i - 1 + while j >= 0: + prev_block_type = page_blocks[j].get("type") + if prev_block_type in available_types: + prev_parent_type = prev_block_type + break + elif prev_block_type == "caption": + # 继续向前查找 + j -= 1 + else: + # 遇到非caption且非table/image的块，停止查找 + break + + # 查找后置位相邻的母块（table或image） + # 向后查找，跳过连续的caption块 + next_parent_type = None + k = i + 1 + while k < n: + next_block_type = page_blocks[k].get("type") + if next_block_type in available_types: + next_parent_type = next_block_type + break + elif next_block_type == "caption": + # 继续向后查找 + k += 1 + else: + # 遇到非caption且非table/image的块，停止查找 + break + + # 根据规则确定caption类型 + new_block = block.copy() + if prev_parent_type: + # 优先使用前置位母块的类型 + new_block["type"] = f"{prev_parent_type}_caption" + elif next_parent_type: + # 没有前置位母块，使用后置位母块的类型 + new_block["type"] = f"{next_parent_type}_caption" + else: + # 没有相邻母块，变更为text + new_block["type"] = "text" + + result_blocks.append(new_block) + + return result_blocks