Sanjayraju30 commited on
Commit
6257859
·
verified ·
1 Parent(s): c925f8d

Update ocr_engine.py

Browse files
Files changed (1) hide show
  1. ocr_engine.py +15 -9
ocr_engine.py CHANGED
@@ -13,17 +13,12 @@ def enhance_image(img):
13
  scale = max_dim / max(height, width)
14
  img = cv2.resize(img, None, fx=scale, fy=scale, interpolation=cv2.INTER_AREA)
15
 
16
- # Grayscale
17
  gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
18
-
19
- # Denoise
20
  gray = cv2.fastNlMeansDenoising(gray, h=15)
21
 
22
- # Sharpen
23
- kernel = np.array([[0, -1, 0], [-1, 5,-1], [0, -1, 0]])
24
  sharp = cv2.filter2D(gray, -1, kernel)
25
 
26
- # Contrast
27
  clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
28
  contrast = clahe.apply(sharp)
29
 
@@ -41,18 +36,29 @@ def extract_weight_from_image(pil_img):
41
  weight_candidates = []
42
 
43
  for _, text, conf in results:
44
- cleaned = text.lower()
 
 
 
 
 
 
45
  cleaned = cleaned.replace("kg", "").replace("kgs", "")
46
- cleaned = cleaned.replace("o", "0").replace("s", "5").replace("g", "9")
47
- cleaned = re.sub(r"[^\d\.]", "", cleaned)
48
 
 
49
  if re.fullmatch(r"\d{2,4}(\.\d{1,2})?", cleaned):
50
  weight_candidates.append((cleaned, conf))
51
 
52
  if not weight_candidates:
53
  return "Not detected", 0.0, "\n".join(ocr_texts)
54
 
 
55
  best_weight, best_conf = sorted(weight_candidates, key=lambda x: -x[1])[0]
 
 
 
 
56
  return best_weight, round(best_conf * 100, 2), "\n".join(ocr_texts)
57
 
58
  except Exception as e:
 
13
  scale = max_dim / max(height, width)
14
  img = cv2.resize(img, None, fx=scale, fy=scale, interpolation=cv2.INTER_AREA)
15
 
 
16
  gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
 
 
17
  gray = cv2.fastNlMeansDenoising(gray, h=15)
18
 
19
+ kernel = np.array([[0, -1, 0], [-1, 5, -1], [0, -1, 0]])
 
20
  sharp = cv2.filter2D(gray, -1, kernel)
21
 
 
22
  clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
23
  contrast = clahe.apply(sharp)
24
 
 
36
  weight_candidates = []
37
 
38
  for _, text, conf in results:
39
+ cleaned = text.lower().strip()
40
+
41
+ # Fix common misreads
42
+ cleaned = cleaned.replace(",", ".") # comma → dot
43
+ cleaned = cleaned.replace("o", "0").replace("O", "0")
44
+ cleaned = cleaned.replace("s", "5").replace("S", "5")
45
+ cleaned = cleaned.replace("g", "9").replace("G", "6")
46
  cleaned = cleaned.replace("kg", "").replace("kgs", "")
47
+ cleaned = re.sub(r"[^\d\.]", "", cleaned) # Keep only digits + dot
 
48
 
49
+ # Match: 2 to 4 digits, optional .digit
50
  if re.fullmatch(r"\d{2,4}(\.\d{1,2})?", cleaned):
51
  weight_candidates.append((cleaned, conf))
52
 
53
  if not weight_candidates:
54
  return "Not detected", 0.0, "\n".join(ocr_texts)
55
 
56
+ # Get highest confidence result
57
  best_weight, best_conf = sorted(weight_candidates, key=lambda x: -x[1])[0]
58
+
59
+ # Remove leading zeros
60
+ best_weight = best_weight.lstrip('0') or '0'
61
+
62
  return best_weight, round(best_conf * 100, 2), "\n".join(ocr_texts)
63
 
64
  except Exception as e: