Update main.py

main.py

@@ -1,33 +1,81 @@
 from fastapi import FastAPI, UploadFile, File
 from paddleocr import PaddleOCR
-from PIL import Image
+from PIL import Image, ImageOps
 import numpy as np
 import io
+import cv2
 
 app = FastAPI()
 
-# Loading the model (Stable 2.7.3 syntax)
-ocr = PaddleOCR(use_angle_cls=False, lang='en', use_gpu=False)
+# ---------------------------------------------------------
+# 🧠 LOAD MODEL - SERVER GRADE CONFIGURATION
+# ---------------------------------------------------------
+# We enable 'use_angle_cls' because we have 16GB RAM now.
+# We keep lang='en' which downloads the standard models.
+ocr = PaddleOCR(
+    use_angle_cls=True,          # ENABLED: Fixes rotation (Crucial for accuracy)
+    lang='en', 
+    use_gpu=False,               # HF Free tier is CPU only
+    show_log=False,
+    
+    # --- ACCURACY TUNING PARAMETERS ---
+    det_db_score_mode='slow',    # SLOWER BUT MORE PRECISE: detailed polygon check
+    det_db_box_thresh=0.5,       # LOWER THRESHOLD: Detects fainter text
+    det_db_unclip_ratio=1.6,     # LARGER BOXES: Prevents cutting off edges of letters
+    cls_thresh=0.9,              # STRICTER ROTATION: Only rotate if 90% sure
+    use_mp=True,                 # MULTI-PROCESSING: Use all CPU cores
+    total_process_num=2          # 2 vCPUs available on HF free tier
+)
 
 @app.get("/")
 def home():
-    return {"status": "OCR Ready"}
+    return {"status": "High-Precision OCR Ready"}
+
+def preprocess_image(image: Image.Image) -> np.ndarray:
+    """
+    Upscales and cleans image for maximum OCR readability.
+    """
+    # 1. Convert to RGB to ensure standard format
+    img = image.convert("RGB")
+    
+    # 2. Upscale small images (OCR hates small text)
+    # If width < 2000px, double the size
+    w, h = img.size
+    if w < 2000:
+        new_w = int(w * 2)
+        new_h = int(h * 2)
+        img = img.resize((new_w, new_h), Image.Resampling.LANCZOS)
+    
+    # 3. Add a white border (padding)
+    # OCR fails if text touches the very edge of the image
+    img = ImageOps.expand(img, border=50, fill='white')
+    
+    return np.array(img)
 
 @app.post("/ocr")
 async def get_ocr(file: UploadFile = File(...)):
     try:
+        # Read image
         content = await file.read()
-        image = Image.open(io.BytesIO(content)).convert("RGB")
-        img_array = np.array(image)
+        pil_image = Image.open(io.BytesIO(content))
+        
+        # --- PRE-PROCESSING STEP ---
+        # Make the image bigger and cleaner
+        img_array = preprocess_image(pil_image)
         
-        # Run OCR
-        result = ocr.ocr(img_array, cls=False)
+        # Run OCR with Classifier Enabled
+        result = ocr.ocr(img_array, cls=True)
         
+        # Extract text
         full_text = ""
+        # Paddle returns a list of lines. If result is None, image was empty.
         if result and result[0]:
+            # result[0] is the list of [box, (text, score)]
+            # We just want the text
             text_lines = [line[1][0] for line in result[0]]
             full_text = "\n".join(text_lines)
         
         return {"text": full_text}
+        
     except Exception as e:
         return {"text": "", "error": str(e)}