Update main.py

main.py

@@ -1,77 +1,41 @@
 from fastapi import FastAPI, UploadFile, File
 from paddleocr import PaddleOCR
-from PIL import Image, ImageOps
+from PIL import Image
 import numpy as np
 import io
-import cv2
 
 app = FastAPI()
 
 # ---------------------------------------------------------
-# 🧠 LOAD MODEL - SERVER GRADE CONFIGURATION
+# 🧠 LOAD MODEL - STANDARD CONFIGURATION
 # ---------------------------------------------------------
-# We enable 'use_angle_cls' because we have 16GB RAM now.
-# We keep lang='en' which downloads the standard models.
+# We use the defaults here because they are generally more robust
+# for standard invoices than the aggressive "High Precision" settings.
 ocr = PaddleOCR(
-    use_angle_cls=True,          # ENABLED: Fixes rotation (Crucial for accuracy)
+    use_angle_cls=True,  # Keep this True to handle rotated pages
     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
+    use_gpu=False
 )
 
 @app.get("/")
 def home():
-    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)
+    return {"status": "Standard OCR Ready"}
 
 @app.post("/ocr")
 async def get_ocr(file: UploadFile = File(...)):
     try:
-        # Read image
+        # 1. Read image
         content = await file.read()
-        pil_image = Image.open(io.BytesIO(content))
-        
-        # --- PRE-PROCESSING STEP ---
-        # Make the image bigger and cleaner
-        img_array = preprocess_image(pil_image)
+        image = Image.open(io.BytesIO(content)).convert("RGB")
+        img_array = np.array(image)
         
-        # Run OCR with Classifier Enabled
+        # 2. Run OCR (Standard Mode)
+        # cls=True ensures we check for rotation
         result = ocr.ocr(img_array, cls=True)
         
-        # Extract text
+        # 3. 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)