feat: add padding

app/api/v1/endpoints/image.py

@@ -2,6 +2,8 @@
 
 import time
 import uuid
+import cv2
+from io import BytesIO
 
 from fastapi import APIRouter, Depends, HTTPException, Request, Response
 
@@ -12,12 +14,15 @@ from app.core.dependencies import (
     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()
 
@@ -63,12 +68,18 @@ async def process_image_ocr(
             image_url=request.image_url,
             image_base64=request.image_base64,
         )
+
+        # Apply padding if enabled (before 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 preprocessing completed in {preprocess_time:.3f}s")
 
-        # Layout detection
+        # Layout detection (using padded image if padding is enabled)
         layout_start = time.time()
-        layout_info = layout_detector.detect(image)
+        layout_info = layout_detector.detect(processed_image)
         layout_time = time.time() - layout_start
         log.info(f"Layout detection completed in {layout_time:.3f}s")
 
@@ -77,11 +88,19 @@ async def process_image_ocr(
         if layout_info.MixedRecognition:
             recognition_method = "MixedRecognition (MinerU)"
             log.info(f"Using {recognition_method}")
-            ocr_result = mineru_service.recognize(image)
+
+            # Convert numpy array to image bytes (image already padded)
+            success, encoded_image = cv2.imencode(".png", processed_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}")
-            ocr_result = glmocr_service.recognize(image)
+            ocr_result = glmocr_service.recognize(processed_image)
         ocr_time = time.time() - ocr_start
 
         total_time = time.time() - preprocess_start