feat: 扫描版PDF支持视觉模型OCR识别
- 添加 pdf_to_images 将PDF页面转为图像 - 添加 extract_text_from_image 使用视觉模型OCR识别图像文字 - 检测扫描版PDF自动切换OCR模式 - glm-4.6v 等视觉模型可识别图像中的文字 - 进度提示显示OCR识别过程
This commit is contained in:
152
services.py
152
services.py
@@ -7,10 +7,20 @@ import json
|
||||
import time
|
||||
import hashlib
|
||||
import threading
|
||||
import base64
|
||||
import io
|
||||
from datetime import datetime, timedelta
|
||||
from pypdf import PdfReader
|
||||
from openai import OpenAI
|
||||
from flask import current_app
|
||||
from PIL import Image
|
||||
|
||||
# pdf2image 用于将PDF转为图像
|
||||
try:
|
||||
from pdf2image import convert_from_path
|
||||
PDF_TO_IMAGE_AVAILABLE = True
|
||||
except ImportError:
|
||||
PDF_TO_IMAGE_AVAILABLE = False
|
||||
|
||||
# ==================== LLM客户端 ====================
|
||||
class TranslationService:
|
||||
@@ -99,6 +109,115 @@ class TranslationService:
|
||||
text = re.sub(r'[\x00-\x08\x0b\x0c\x0e-\x1f]', '', text)
|
||||
return text.strip()
|
||||
|
||||
def is_vision_model(self):
|
||||
"""检查是否是视觉模型"""
|
||||
model = self.llm_config.get('model', '')
|
||||
# 常见视觉模型名称
|
||||
vision_models = ['vision', 'vlm', 'glm-4v', 'glm-4.6v', 'gpt-4-vision', 'gpt-4o', 'qwen-vl', 'claude-3']
|
||||
return any(v in model.lower() for v in vision_models)
|
||||
|
||||
def pdf_to_images(self, pdf_path, max_pages=None):
|
||||
"""将PDF页面转换为图像"""
|
||||
if not PDF_TO_IMAGE_AVAILABLE:
|
||||
return None, "pdf2image未安装，无法处理扫描版PDF。请安装: pip install pdf2image"
|
||||
|
||||
try:
|
||||
# 获取PDF页数
|
||||
reader = PdfReader(pdf_path)
|
||||
total_pages = len(reader.pages)
|
||||
|
||||
if max_pages:
|
||||
pages_to_convert = min(max_pages, total_pages)
|
||||
else:
|
||||
pages_to_convert = total_pages
|
||||
|
||||
# 转换PDF为图像
|
||||
images = convert_from_path(
|
||||
pdf_path,
|
||||
first_page=1,
|
||||
last_page=pages_to_convert,
|
||||
dpi=200, # 适当的DPI
|
||||
fmt='jpeg'
|
||||
)
|
||||
|
||||
return images, None
|
||||
|
||||
except Exception as e:
|
||||
return None, f"PDF转图像失败: {str(e)}"
|
||||
|
||||
def extract_text_from_image(self, image):
|
||||
"""使用视觉模型从图像中提取文字"""
|
||||
if not self.is_vision_model():
|
||||
return None, "当前模型不是视觉模型，无法识别图像文字"
|
||||
|
||||
try:
|
||||
# 将图像转为base64
|
||||
buffered = io.BytesIO()
|
||||
image.save(buffered, format="JPEG")
|
||||
img_base64 = base64.b64encode(buffered.getvalue()).decode('utf-8')
|
||||
|
||||
# 构建多模态请求
|
||||
response = self.client.chat.completions.create(
|
||||
model=self.llm_config['model'],
|
||||
messages=[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "text",
|
||||
"text": "请识别并提取这张图片中的所有文字内容。只输出提取的文字，不要添加任何解释或说明。保持原有的段落和格式。"
|
||||
},
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {
|
||||
"url": f"data:image/jpeg;base64,{img_base64}"
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
],
|
||||
max_tokens=self.llm_config['max_tokens'],
|
||||
temperature=0.1,
|
||||
timeout=self.llm_config['timeout'],
|
||||
)
|
||||
|
||||
content = response.choices[0].message.content
|
||||
return content.strip() if content else '', None
|
||||
|
||||
except Exception as e:
|
||||
return '', f"视觉模型识别失败: {str(e)}"
|
||||
|
||||
def extract_text_from_scanned_pdf(self, pdf_path, progress_callback=None):
|
||||
"""从扫描版PDF提取文字（使用视觉模型OCR）"""
|
||||
images, error = self.pdf_to_images(pdf_path)
|
||||
|
||||
if error:
|
||||
return [], error
|
||||
|
||||
pages_text = []
|
||||
total = len(images)
|
||||
|
||||
for i, image in enumerate(images):
|
||||
if progress_callback:
|
||||
progress_callback(int((i+1)/total*50), total, f"OCR识别第{i+1}页...")
|
||||
|
||||
text, err = self.extract_text_from_image(image)
|
||||
|
||||
if err:
|
||||
pages_text.append({
|
||||
'page': i + 1,
|
||||
'text': '',
|
||||
'error': err
|
||||
})
|
||||
else:
|
||||
pages_text.append({
|
||||
'page': i + 1,
|
||||
'text': text or '',
|
||||
'error': None
|
||||
})
|
||||
|
||||
return pages_text, None
|
||||
|
||||
def chunk_text(self, text, max_size=2000):
|
||||
"""分块"""
|
||||
paragraphs = text.split('\n\n')
|
||||
@@ -131,19 +250,38 @@ class TranslationService:
|
||||
Returns:
|
||||
翻译统计信息
|
||||
"""
|
||||
# 先尝试常规提取
|
||||
pages = self.extract_pdf_text(pdf_path)
|
||||
total_pages = len(pages)
|
||||
|
||||
# 检查是否有可翻译内容
|
||||
total_text = sum(len(p['text']) for p in pages)
|
||||
|
||||
# 如果无法提取文本，尝试使用视觉模型OCR
|
||||
if total_pages == 0 or total_text < 10:
|
||||
error_msg = "PDF无法提取文本内容。可能原因：\n1. PDF是扫描版（图像形式），需要OCR处理\n2. PDF为空或加密\n请使用包含可提取文本的PDF文件。"
|
||||
if self.is_vision_model() and PDF_TO_IMAGE_AVAILABLE:
|
||||
if progress_callback:
|
||||
progress_callback(0, 0, "检测到扫描版PDF，使用视觉模型OCR...")
|
||||
|
||||
pages, error = self.extract_text_from_scanned_pdf(pdf_path, progress_callback)
|
||||
|
||||
if error:
|
||||
raise ValueError(error)
|
||||
|
||||
total_pages = len(pages)
|
||||
total_text = sum(len(p['text']) for p in pages)
|
||||
|
||||
if total_text < 10:
|
||||
raise ValueError("视觉模型OCR未能提取到有效文字内容")
|
||||
|
||||
if progress_callback:
|
||||
progress_callback(50, total_pages, "OCR完成，开始翻译...")
|
||||
else:
|
||||
error_msg = "PDF无法提取文本内容。可能原因：\n1. PDF是扫描版（图像形式）\n2. 当前大模型不是视觉模型，无法识别图像文字\n\n如需处理扫描版PDF，请配置视觉大模型（如 glm-4.6v、gpt-4-vision）"
|
||||
if progress_callback:
|
||||
progress_callback(0, 0, error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
if progress_callback:
|
||||
progress_callback(0, total_pages, "开始翻译...")
|
||||
progress_callback(50, total_pages, "开始翻译...")
|
||||
|
||||
translated_pages = []
|
||||
total_chunks = 0
|
||||
@@ -158,8 +296,10 @@ class TranslationService:
|
||||
translated_chunks.append(translated)
|
||||
|
||||
if progress_callback:
|
||||
progress = int((i + 1) / len(chunks) * 100 / total_pages)
|
||||
progress_callback(progress, total_pages, f"翻译第{page_data['page']}页")
|
||||
# OCR占50%，翻译占50%
|
||||
page_progress = (i + 1) / len(chunks)
|
||||
overall_progress = 50 + int(page_progress * 50 / total_pages)
|
||||
progress_callback(overall_progress, total_pages, f"翻译第{page_data['page']}页")
|
||||
|
||||
translated_pages.append({
|
||||
'page': page_data['page'],
|
||||
|
||||
Reference in New Issue
Block a user