feat(ocr): integrate preprocessing into stream and batch endpoints

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

ocr-service/main.py

@@ -25,6 +25,7 @@ from confidence import apply_confidence_markers, get_threshold
 from engines import kraken as kraken_engine
 from engines import surya as surya_engine
 from models import OcrBlock, OcrRequest
+from preprocessing import preprocess_page
 
 TRAINING_TOKEN = os.environ.get("TRAINING_TOKEN", "")
 KRAKEN_MODEL_PATH = os.environ.get("KRAKEN_MODEL_PATH", "/app/models/german_kurrent.mlmodel")
@@ -86,6 +87,10 @@ async def run_ocr(request: OcrRequest):
 
     images = await _download_and_convert_pdf(request.pdfUrl)
 
+    for i, img in enumerate(images):
+        images[i] = await asyncio.to_thread(preprocess_page, img)
+        del img
+
     script_type = request.scriptType.upper()
 
     if script_type == "HANDWRITING_KURRENT":
@@ -157,6 +162,8 @@ async def run_ocr_stream(request: OcrRequest):
                     continue
 
                 try:
+                    yield json.dumps({"type": "preprocessing", "pageNumber": page_idx}) + "\n"
+                    image = await asyncio.to_thread(preprocess_page, image)
                     blocks = []
                     for region in page_regions:
                         text = await asyncio.to_thread(
@@ -214,6 +221,8 @@ async def run_ocr_stream(request: OcrRequest):
 
         for page_idx, image in enumerate(images, start=1):
             try:
+                yield json.dumps({"type": "preprocessing", "pageNumber": page_idx}) + "\n"
+                image = await asyncio.to_thread(preprocess_page, image)
                 blocks = await asyncio.to_thread(
                     engine.extract_page_blocks, image, page_idx, request.language
                 )