Files
doc_processer/app/api/v1/endpoints/image.py
2026-02-07 21:38:41 +08:00

144 lines
5.7 KiB
Python

"""Image OCR endpoint."""
import time
import uuid
import cv2
from io import BytesIO
from fastapi import APIRouter, Depends, HTTPException, Request, Response
from app.core.dependencies import (
get_image_processor,
get_layout_detector,
get_ocr_service,
get_mineru_ocr_service,
get_glmocr_service,
)
from app.core.config import get_settings
from app.core.logging_config import get_logger, RequestIDAdapter
from app.schemas.image import ImageOCRRequest, ImageOCRResponse
from app.services.image_processor import ImageProcessor
from app.services.layout_detector import LayoutDetector
from app.services.ocr_service import OCRService, MineruOCRService, GLMOCRService
settings = get_settings()
router = APIRouter()
logger = get_logger()
@router.post("/ocr", response_model=ImageOCRResponse)
async def process_image_ocr(
request: ImageOCRRequest,
http_request: Request,
response: Response,
image_processor: ImageProcessor = Depends(get_image_processor),
layout_detector: LayoutDetector = Depends(get_layout_detector),
mineru_service: MineruOCRService = Depends(get_mineru_ocr_service),
paddle_service: OCRService = Depends(get_ocr_service),
glmocr_service: GLMOCRService = Depends(get_glmocr_service),
) -> ImageOCRResponse:
"""Process an image and extract content as LaTeX, Markdown, and MathML.
The processing pipeline:
1. Load and preprocess image (add 30% whitespace padding)
2. Detect layout using DocLayout-YOLO
3. Based on layout:
- If plain text exists: use PP-DocLayoutV2 for mixed recognition
- Otherwise: use PaddleOCR-VL with formula prompt
4. Convert output to LaTeX, Markdown, and MathML formats
Note: OMML conversion is not included due to performance overhead.
Use the /convert/latex-to-omml endpoint to convert LaTeX to OMML separately.
"""
# Get or generate request ID
request_id = http_request.headers.get("x-request-id", str(uuid.uuid4()))
response.headers["x-request-id"] = request_id
# Create logger adapter with request_id
log = RequestIDAdapter(logger, {"request_id": request_id})
log.request_id = request_id
try:
log.info("Starting image OCR processing")
# Preprocess image (load only, no padding yet)
preprocess_start = time.time()
image = image_processor.preprocess(
image_url=request.image_url,
image_base64=request.image_base64,
)
# Apply padding only for layout detection
processed_image = image
if image_processor and settings.is_padding:
processed_image = image_processor.add_padding(image)
preprocess_time = time.time() - preprocess_start
log.debug(f"Image loading completed in {preprocess_time:.3f}s")
# Layout detection (using padded image if padding is enabled)
layout_start = time.time()
layout_info = layout_detector.detect(processed_image)
layout_time = time.time() - layout_start
log.info(f"Layout detection completed in {layout_time:.3f}s")
# OCR recognition (use original image without padding)
ocr_start = time.time()
if layout_info.MixedRecognition:
recognition_method = "MixedRecognition (MinerU)"
log.info(f"Using {recognition_method}")
# Convert original image (without padding) to bytes
success, encoded_image = cv2.imencode(".png", image)
if not success:
raise RuntimeError("Failed to encode image")
image_bytes = BytesIO(encoded_image.tobytes())
image_bytes.seek(0) # Ensure position is at the beginning
ocr_result = mineru_service.recognize(image_bytes)
else:
recognition_method = "FormulaOnly (GLMOCR)"
log.info(f"Using {recognition_method}")
# Try GLM-OCR first, fallback to MinerU if token limit exceeded
try:
ocr_result = glmocr_service.recognize(image)
except Exception as e:
error_msg = str(e)
# Check if error is due to token limit (max_model_len exceeded)
if "max_model_len" in error_msg or "decoder prompt" in error_msg or "BadRequestError" in error_msg:
log.warning(f"GLM-OCR failed due to token limit: {error_msg}")
log.info("Falling back to MinerU for recognition")
recognition_method = "FormulaOnly (MinerU fallback)"
# Convert original image to bytes for MinerU
success, encoded_image = cv2.imencode(".png", image)
if not success:
raise RuntimeError("Failed to encode image")
image_bytes = BytesIO(encoded_image.tobytes())
image_bytes.seek(0)
ocr_result = mineru_service.recognize(image_bytes)
else:
# Re-raise other errors
raise
ocr_time = time.time() - ocr_start
total_time = time.time() - preprocess_start
log.info(f"OCR processing completed - Method: {recognition_method}, " f"Layout time: {layout_time:.3f}s, OCR time: {ocr_time:.3f}s, " f"Total time: {total_time:.3f}s")
except RuntimeError as e:
log.error(f"OCR processing failed: {str(e)}", exc_info=True)
raise HTTPException(status_code=503, detail=str(e))
except Exception as e:
log.error(f"Unexpected error during OCR processing: {str(e)}", exc_info=True)
raise HTTPException(status_code=500, detail="Internal server error")
return ImageOCRResponse(
latex=ocr_result.get("latex", ""),
markdown=ocr_result.get("markdown", ""),
mathml=ocr_result.get("mathml", ""),
mml=ocr_result.get("mml", ""),
)