Upload main.py

main.py

@@ -7,6 +7,7 @@ import uvicorn
 import os
 import numpy as np
 import cv2
+import re
 
 # PDF support
 try:
@@ -35,6 +36,23 @@ async def startup_event():
     print("Server started. OCR models will be loaded lazily on first request.")
 
 
+# -------------------- تنظيف النص العربي --------------------
+def clean_arabic_text(text: str) -> str:
+    if not text:
+        return ""
+
+    # إزالة أي شيء غير عربي أو أرقام
+    text = re.sub(r"[^\u0600-\u06FF0-9]", "", text)
+
+    # إزالة التشكيل
+    text = re.sub(r"[\u064B-\u065F]", "", text)
+
+    # إزالة أي مسافات
+    text = re.sub(r"\s+", "", text)
+
+    return text.strip()
+
+
 def get_models():
     global paddle_detector, paddle_recognizer
 
@@ -54,10 +72,16 @@ def get_models():
     return paddle_detector, paddle_recognizer
 
 
-def process_image(img: np.ndarray, detector, recognizer, min_conf: float) -> List[Dict]:
+def process_image(
+    img: np.ndarray,
+    detector,
+    recognizer,
+    min_conf: float
+) -> List[Dict]:
+
     h_img, w_img = img.shape[:2]
 
-    # 1️⃣ Detect text boxes
+    # 1️⃣ Detect text
     results = detector.predict(img)
 
     all_rois = []
@@ -78,7 +102,7 @@ def process_image(img: np.ndarray, detector, recognizer, min_conf: float) -> Lis
                 roi = img[y1:y2, x1:x2]
                 if roi.size > 0:
                     all_rois.append(roi)
-                    all_bboxes.append([int(x1), int(y1), int(x2), int(y2)])
+                    all_bboxes.append([x1, y1, x2, y2])
 
     # 2️⃣ Recognize text
     ocr_results = []
@@ -87,13 +111,14 @@ def process_image(img: np.ndarray, detector, recognizer, min_conf: float) -> Lis
         try:
             rec_gen = recognizer.predict(roi)
             rec = next(rec_gen)
-            text = rec.get("rec_text", "")
+            raw_text = rec.get("rec_text", "")
             score = float(rec.get("rec_score", 0.0))
+            text = clean_arabic_text(raw_text)
         except:
             text = ""
             score = 0.0
 
-        if score >= min_conf and text.strip():
+        if score >= min_conf and text:
             ocr_results.append({
                 "box_id": i + 1,
                 "text": text,
@@ -101,12 +126,11 @@ def process_image(img: np.ndarray, detector, recognizer, min_conf: float) -> Lis
                 "bbox": all_bboxes[i]
             })
 
-    # ✅ الحل الأساسي هنا
-    # ترتيب عربي: من فوق لتحت ثم من اليمين لليسار
+    # ✅ ترتيب عربي: فوق → تحت ، يمين → شمال
     ocr_results.sort(
         key=lambda x: (
-            x["bbox"][1],     # Y (top → bottom)
-            -x["bbox"][0]     # X (right → left)
+            x["bbox"][1],     # Y
+            -x["bbox"][0]     # X (RTL)
         )
     )