ax-backend-v1/app/services/image_service.py

"""image_service.py — 图片提取和处理服务"""

import base64
import json
from typing import Dict, List

from docx import Document
from docx.enum.text import WD_ALIGN_PARAGRAPH


def extract_images_from_word(doc: Document) -> List[Dict]:
    """从 Word 文档提取图片及基本样式信息
    
    Args:
        doc: python-docx Document 对象
        
    Returns:
        图片列表，每个元素包含：
        - paragraph_index: 图片所在段落索引
        - data_url: Base64 编码的 Data URL
        - style: 样式信息（宽度、高度、对齐方式、段落样式）
    """
    images = []
    
    # 建立 rel_id -> 图片数据映射
    image_parts = {}
    for rel in doc.part.rels.values():
        if "image" in rel.target_ref:
            image_parts[rel.rId] = {
                "blob": rel.target_part.blob,
                "content_type": rel.target_part.content_type
            }
    
    # 遍历段落查找图片
    for para_idx, paragraph in enumerate(doc.paragraphs):
        for run in paragraph.runs:
            # 查找 inline 图片（w:drawing 元素）
            for drawing in run._element.findall('.//{http://schemas.openxmlformats.org/wordprocessingml/2006/main}drawing'):
                # 提取图片引用
                blip = drawing.find('.//{http://schemas.openxmlformats.org/drawingml/2006/main}blip')
                if blip is None:
                    continue
                    
                rel_id = blip.get('{http://schemas.openxmlformats.org/officeDocument/2006/relationships}embed')
                if not rel_id or rel_id not in image_parts:
                    continue
                
                # 提取尺寸（EMU 转厘米，1厘米 = 360000 EMU）
                extent = drawing.find('.//{http://schemas.openxmlformats.org/drawingml/2006/wordprocessingDrawing}extent')
                width_emu = int(extent.get('cx')) if extent is not None else 360000 * 10  # 默认 10cm
                height_emu = int(extent.get('cy')) if extent is not None else 360000 * 7   # 默认 7cm
                
                width_cm = round(width_emu / 360000, 2)
                height_cm = round(height_emu / 360000, 2)
                
                # 获取对齐方式
                align = 'left'
                if paragraph.alignment == WD_ALIGN_PARAGRAPH.CENTER:
                    align = 'center'
                elif paragraph.alignment == WD_ALIGN_PARAGRAPH.RIGHT:
                    align = 'right'
                
                # 获取段落样式名称
                para_style = paragraph.style.name if paragraph.style else "Normal"
                
                # Base64 编码
                image_blob = image_parts[rel_id]['blob']
                content_type = image_parts[rel_id]['content_type']
                b64_data = base64.b64encode(image_blob).decode('utf-8')
                data_url = f"data:{content_type};base64,{b64_data}"
                
                images.append({
                    "paragraph_index": para_idx,
                    "data_url": data_url,
                    "style": {
                        "width": width_cm,
                        "height": height_cm,
                        "align": align,
                        "unit": "cm",
                        "para_style": para_style  # 新增：段落样式
                    }
                })
    
    return images


def create_image_markdown(data_url: str, style: Dict, alt: str = "图片") -> str:
    """生成带样式注释的图片 Markdown
    
    Args:
        data_url: Base64 编码的 Data URL
        style: 样式字典（width, height, align）
        alt: 图片替代文本
        
    Returns:
        格式化的 Markdown 字符串
        
    Example:
        <!-- img-style: {"width": 4.0, "height": 3.0, "align": "center"} -->
        ![图片](data:image/png;base64,...)
    """
    style_json = json.dumps(style, ensure_ascii=False)
    return f'<!-- img-style: {style_json} -->\n![{alt}]({data_url})\n'