我正在使用Docling库将PDF文件内容提取并转换为HTML格式，以保留原文档的结构与格式，使用的代码如下：

from docling.document_converter import DocumentConverter, PdfFormatOption
from docling.datamodel.pipeline_options import PdfPipelineOptions
from docling.datamodel.base_models import InputFormat
from docling_core.types.doc import ImageRefMode
from pathlib import Path
import logging

# Set up logging
logging.basicConfig(level=logging.INFO)
log = logging.getLogger(__name__)  # Corrected: _name_ -> __name__

# Configure image settings
IMAGE_RESOLUTION_SCALE = 2.0

# Path to your PDF file
source = Path(r"C:\Users\Downloads\Journal.pdf")
output_path = Path(r"C:\Users\Desktop\output20.html")

# Configure pipeline options for image handling
pipeline_options = PdfPipelineOptions()
pipeline_options.images_scale = IMAGE_RESOLUTION_SCALE
pipeline_options.generate_page_images = True
pipeline_options.generate_picture_images = True

# Create converter instance with image options
converter = DocumentConverter(
    format_options={
        InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)
    }
)

# Convert PDF to document
result = converter.convert(source)

# Save HTML with embedded images
result.document.save_as_html(output_path, image_mode=ImageRefMode.EMBEDDED)

log.info(f"HTML file with embedded images created at: {output_path}")

目前转换后的HTML内容顺序与原PDF一致，但表格结构的处理存在以下问题：

• 表格未被正确识别，表格数据以<p>标签呈现而非标准的<table>、<tr>、<td>结构；
• 表格内容错位，单元格内容被错误拆分，且单元格内的图片被移至表格外；
• 含嵌入图片的复杂表格无法被正确保留。

我的预期效果是：提取PDF表格时保留正确的行列结构，以标准HTML表格标签呈现，同时保留原格式、对齐方式及单元格内容。

我已尝试的操作：检查HTML输出、调整PdfPipelineOptions参数、对比Document Intelligence库（该库页眉页脚提取更优，但复杂表格仍处理不佳）。

现寻求解决方案：如何通过Docling库正确提取PDF表格并转换为保留原布局与格式的结构化HTML？

1. 启用表格检测与结构化核心配置 #

Docling默认未完全开启结构化表格提取，需在PdfPipelineOptions中添加表格处理相关参数：

pipeline_options = PdfPipelineOptions()
# 开启表格检测功能
pipeline_options.enable_table_detection = True
# 设置结构化表格识别模式，优先生成标准HTML表格标签
pipeline_options.table_recognition_mode = "STRUCTURED"
# 保留单元格内的格式与图片位置
pipeline_options.preserve_cell_content_formatting = True
# 通过字体分析辅助识别表格边界
pipeline_options.enable_font_analysis = True
# 原有图片配置
pipeline_options.images_scale = IMAGE_RESOLUTION_SCALE
pipeline_options.generate_page_images = True
pipeline_options.generate_picture_images = True

2. 调整HTML保存时的表格渲染策略 #

调用save_as_html时，强制指定结构化表格渲染模式，避免降级为流式文本：

result.document.save_as_html(
    output_path,
    image_mode=ImageRefMode.EMBEDDED,
    render_tables_as_structured=True
)

3. 完整调整后的代码 #

from docling.document_converter import DocumentConverter, PdfFormatOption
from docling.datamodel.pipeline_options import PdfPipelineOptions
from docling.datamodel.base_models import InputFormat
from docling_core.types.doc import ImageRefMode
from pathlib import Path
import logging

# Set up logging
logging.basicConfig(level=logging.INFO)
log = logging.getLogger(__name__)

# Configure image settings
IMAGE_RESOLUTION_SCALE = 2.5  # 提升分辨率帮助精准识别表格

# Path to your PDF file
source = Path(r"C:\Users\Downloads\Journal.pdf")
output_path = Path(r"C:\Users\Desktop\output20.html")

# Configure pipeline options for table and image handling
pipeline_options = PdfPipelineOptions()
# 表格处理核心配置
pipeline_options.enable_table_detection = True
pipeline_options.table_recognition_mode = "STRUCTURED"
pipeline_options.preserve_cell_content_formatting = True
pipeline_options.enable_font_analysis = True
# 图片配置
pipeline_options.images_scale = IMAGE_RESOLUTION_SCALE
pipeline_options.generate_page_images = True
pipeline_options.generate_picture_images = True

# Create converter instance with updated options
converter = DocumentConverter(
    format_options={
        InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)
    }
)

# Convert PDF to document
result = converter.convert(source)

# Save HTML with structured tables and embedded images
result.document.save_as_html(
    output_path,
    image_mode=ImageRefMode.EMBEDDED,
    render_tables_as_structured=True
)

log.info(f"HTML file with structured tables created at: {output_path}")

参数说明 #

• enable_table_detection：触发Docling对PDF中表格区域的识别；
• table_recognition_mode = "STRUCTURED"：强制以行列结构解析表格，生成标准<table>系列标签；
• preserve_cell_content_formatting：保留单元格内的文本样式、图片位置，避免内容错位或图片被移出表格；
• enable_font_analysis：通过字体大小、样式差异辅助判断表格边界与单元格范围；
• render_tables_as_structured=True：确保HTML输出时直接渲染结构化表格，而非将表格内容转为普通段落。

内容的提问来源于stack exchange，提问作者Akshata