Sanjayraju30 commited on
Commit
292bc54
·
verified ·
1 Parent(s): 2f2d5f6

Update ocr_engine.py

Browse files
Files changed (1) hide show
  1. ocr_engine.py +6 -9
ocr_engine.py CHANGED
@@ -10,14 +10,14 @@ def extract_weight_from_image(pil_img):
10
  try:
11
  img = np.array(pil_img)
12
 
13
- # Resize very large images
14
  max_dim = 1000
15
  height, width = img.shape[:2]
16
  if max(height, width) > max_dim:
17
  scale = max_dim / max(height, width)
18
  img = cv2.resize(img, None, fx=scale, fy=scale, interpolation=cv2.INTER_AREA)
19
 
20
- # OCR recognition
21
  results = reader.readtext(img)
22
  print("DEBUG OCR RESULTS:", results)
23
 
@@ -26,11 +26,10 @@ def extract_weight_from_image(pil_img):
26
  fallback_weight = None
27
  fallback_conf = 0.0
28
 
29
- for _, (text, conf) in results:
30
  original = text
31
  cleaned = text.lower().strip()
32
 
33
- # Fix common OCR misreads
34
  cleaned = cleaned.replace(",", ".")
35
  cleaned = cleaned.replace("o", "0").replace("O", "0")
36
  cleaned = cleaned.replace("s", "5").replace("S", "5")
@@ -40,24 +39,22 @@ def extract_weight_from_image(pil_img):
40
 
41
  raw_texts.append(f"{original} → {cleaned} (conf: {round(conf, 2)})")
42
 
43
- # Save fallback if no match later
44
  if cleaned and cleaned.replace(".", "").isdigit() and not fallback_weight:
45
  fallback_weight = cleaned
46
  fallback_conf = conf
47
 
48
- # Match proper weight format: 75.02, 97.2, 105
49
  if cleaned.count(".") <= 1 and re.fullmatch(r"\d{2,4}(\.\d{1,3})?", cleaned):
50
  weight_candidates.append((cleaned, conf))
51
 
52
- # Choose best candidate
53
  if weight_candidates:
54
  best_weight, best_conf = sorted(weight_candidates, key=lambda x: -x[1])[0]
55
  elif fallback_weight:
56
  best_weight, best_conf = fallback_weight, fallback_conf
57
  else:
58
- return "Not detected", 0.0, "\n".join(raw_texts)
59
 
60
- # Strip unnecessary leading zeros
61
  if "." in best_weight:
62
  int_part, dec_part = best_weight.split(".")
63
  int_part = int_part.lstrip("0") or "0"
 
10
  try:
11
  img = np.array(pil_img)
12
 
13
+ # Resize if large
14
  max_dim = 1000
15
  height, width = img.shape[:2]
16
  if max(height, width) > max_dim:
17
  scale = max_dim / max(height, width)
18
  img = cv2.resize(img, None, fx=scale, fy=scale, interpolation=cv2.INTER_AREA)
19
 
20
+ # OCR
21
  results = reader.readtext(img)
22
  print("DEBUG OCR RESULTS:", results)
23
 
 
26
  fallback_weight = None
27
  fallback_conf = 0.0
28
 
29
+ for _, (text, conf) in results: # ✅ Correct unpacking
30
  original = text
31
  cleaned = text.lower().strip()
32
 
 
33
  cleaned = cleaned.replace(",", ".")
34
  cleaned = cleaned.replace("o", "0").replace("O", "0")
35
  cleaned = cleaned.replace("s", "5").replace("S", "5")
 
39
 
40
  raw_texts.append(f"{original} → {cleaned} (conf: {round(conf, 2)})")
41
 
 
42
  if cleaned and cleaned.replace(".", "").isdigit() and not fallback_weight:
43
  fallback_weight = cleaned
44
  fallback_conf = conf
45
 
 
46
  if cleaned.count(".") <= 1 and re.fullmatch(r"\d{2,4}(\.\d{1,3})?", cleaned):
47
  weight_candidates.append((cleaned, conf))
48
 
49
+ # Choose result
50
  if weight_candidates:
51
  best_weight, best_conf = sorted(weight_candidates, key=lambda x: -x[1])[0]
52
  elif fallback_weight:
53
  best_weight, best_conf = fallback_weight, fallback_conf
54
  else:
55
+ return "Not detected", 0.0, "OCR returned nothing useful"
56
 
57
+ # Clean leading zeros
58
  if "." in best_weight:
59
  int_part, dec_part = best_weight.split(".")
60
  int_part = int_part.lstrip("0") or "0"