Spaces:
Sleeping
Sleeping
Update ocr_engine.py
Browse files- ocr_engine.py +76 -4
ocr_engine.py
CHANGED
|
@@ -93,7 +93,7 @@ def detect_roi(img):
|
|
| 93 |
logging.info("No suitable ROI found, attempting fallback criteria.")
|
| 94 |
# Fallback with relaxed criteria
|
| 95 |
valid_contours = [c for c in contours if 500 < cv2.contourArea(c) < (img_area * 0.95) and
|
| 96 |
-
|
| 97 |
if valid_contours:
|
| 98 |
contour = max(valid_contours, key=cv2.contourArea)
|
| 99 |
x, y, w, h = cv2.boundingRect(contour)
|
|
@@ -254,11 +254,11 @@ def extract_weight_from_image(pil_img):
|
|
| 254 |
|
| 255 |
brightness = estimate_brightness(img)
|
| 256 |
conf_threshold = 0.7 if brightness > 150 else (0.6 if brightness > 80 else 0.4)
|
| 257 |
-
|
|
|
|
| 258 |
roi_area = roi_bbox[2] * roi_bbox[3]
|
| 259 |
conf_threshold *= 1.2 if roi_area > (img.shape[0] * img.shape[1] * 0.5) else 1.0
|
| 260 |
|
| 261 |
-
roi_img, roi_bbox = detect_roi(img)
|
| 262 |
custom_result = custom_seven_segment_ocr(roi_img, roi_bbox)
|
| 263 |
if custom_result:
|
| 264 |
try:
|
|
@@ -285,4 +285,76 @@ def extract_weight_from_image(pil_img):
|
|
| 285 |
results = easyocr_reader.readtext(final_roi, detail=1, paragraph=False,
|
| 286 |
contrast_ths=0.4, adjust_contrast=1.2,
|
| 287 |
text_threshold=0.5, mag_ratio=4.0,
|
| 288 |
-
allowlist='0123456789. kglb', batch_size=
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
logging.info("No suitable ROI found, attempting fallback criteria.")
|
| 94 |
# Fallback with relaxed criteria
|
| 95 |
valid_contours = [c for c in contours if 500 < cv2.contourArea(c) < (img_area * 0.95) and
|
| 96 |
+
0.8 <= cv2.boundingRect(c)[2]/cv2.boundingRect(c)[3] <= 12.0]
|
| 97 |
if valid_contours:
|
| 98 |
contour = max(valid_contours, key=cv2.contourArea)
|
| 99 |
x, y, w, h = cv2.boundingRect(contour)
|
|
|
|
| 254 |
|
| 255 |
brightness = estimate_brightness(img)
|
| 256 |
conf_threshold = 0.7 if brightness > 150 else (0.6 if brightness > 80 else 0.4)
|
| 257 |
+
roi_img, roi_bbox = detect_roi(img)
|
| 258 |
+
if roi_bbox:
|
| 259 |
roi_area = roi_bbox[2] * roi_bbox[3]
|
| 260 |
conf_threshold *= 1.2 if roi_area > (img.shape[0] * img.shape[1] * 0.5) else 1.0
|
| 261 |
|
|
|
|
| 262 |
custom_result = custom_seven_segment_ocr(roi_img, roi_bbox)
|
| 263 |
if custom_result:
|
| 264 |
try:
|
|
|
|
| 285 |
results = easyocr_reader.readtext(final_roi, detail=1, paragraph=False,
|
| 286 |
contrast_ths=0.4, adjust_contrast=1.2,
|
| 287 |
text_threshold=0.5, mag_ratio=4.0,
|
| 288 |
+
allowlist='0123456789. kglb', batch_size=batch_size, y_ths=0.2)
|
| 289 |
+
|
| 290 |
+
best_weight = None
|
| 291 |
+
best_conf = 0.0
|
| 292 |
+
best_score = 0.0
|
| 293 |
+
unit = None
|
| 294 |
+
for (bbox, text, conf) in results:
|
| 295 |
+
if 'kg' in text.lower():
|
| 296 |
+
unit = 'kg'
|
| 297 |
+
continue
|
| 298 |
+
elif 'g' in text.lower():
|
| 299 |
+
unit = 'g'
|
| 300 |
+
continue
|
| 301 |
+
elif 'lb' in text.lower():
|
| 302 |
+
unit = 'lb'
|
| 303 |
+
continue
|
| 304 |
+
text = re.sub(r"[^\d\.]", "", text)
|
| 305 |
+
if text.count('.') > 1:
|
| 306 |
+
text = text.replace('.', '', text.count('.') - 1)
|
| 307 |
+
text = text.strip('.')
|
| 308 |
+
if re.fullmatch(r"^\d*\.?\d*$", text):
|
| 309 |
+
try:
|
| 310 |
+
weight = float(text)
|
| 311 |
+
if unit == 'g':
|
| 312 |
+
weight /= 1000 # Convert grams to kilograms
|
| 313 |
+
elif unit == 'lb':
|
| 314 |
+
weight *= 0.453592 # Convert pounds to kilograms
|
| 315 |
+
range_score = 1.5 if 0.01 <= weight <= 500 else 0.8
|
| 316 |
+
digit_count = len(text.replace('.', ''))
|
| 317 |
+
digit_score = 1.3 if 2 <= digit_count <= 6 else 0.9
|
| 318 |
+
score = conf * range_score * digit_score
|
| 319 |
+
if roi_bbox:
|
| 320 |
+
(x_roi, y_roi, w_roi, h_roi) = roi_bbox
|
| 321 |
+
roi_area = w_roi * h_roi
|
| 322 |
+
x_min, y_min = int(min(b[0] for b in bbox)), int(min(b[1] for b in bbox))
|
| 323 |
+
x_max, y_max = int(max(b[0] for b in bbox)), int(max(b[1] for b in bbox))
|
| 324 |
+
bbox_area = (x_max - x_min) * (y_max - y_min)
|
| 325 |
+
if roi_area > 0 and bbox_area / roi_area < 0.05:
|
| 326 |
+
score *= 0.6
|
| 327 |
+
if score > best_score and conf > conf_threshold:
|
| 328 |
+
best_weight = text
|
| 329 |
+
best_conf = conf
|
| 330 |
+
best_score = score
|
| 331 |
+
logging.info(f"Candidate EasyOCR weight: '{text}', Unit: {unit or 'none'}, Conf: {conf}, Score: {score}")
|
| 332 |
+
except ValueError:
|
| 333 |
+
logging.warning(f"Could not convert '{text}' to float during EasyOCR fallback.")
|
| 334 |
+
|
| 335 |
+
if not best_weight:
|
| 336 |
+
logging.info("No valid weight detected after all attempts.")
|
| 337 |
+
return "Not detected", 0.0
|
| 338 |
+
|
| 339 |
+
# Format the weight
|
| 340 |
+
if "." in best_weight:
|
| 341 |
+
int_part, dec_part = best_weight.split(".")
|
| 342 |
+
int_part = int_part.lstrip("0") or "0"
|
| 343 |
+
dec_part = dec_part.rstrip('0')
|
| 344 |
+
best_weight = f"{int_part}.{dec_part}" if dec_part else int_part
|
| 345 |
+
else:
|
| 346 |
+
best_weight = best_weight.lstrip('0') or "0"
|
| 347 |
+
|
| 348 |
+
try:
|
| 349 |
+
final_weight = float(best_weight)
|
| 350 |
+
if final_weight < 0.01 or final_weight > 500:
|
| 351 |
+
best_conf *= 0.7
|
| 352 |
+
except ValueError:
|
| 353 |
+
pass
|
| 354 |
+
|
| 355 |
+
logging.info(f"Final detected weight: {best_weight}, Unit: {unit or 'none'}, Confidence: {round(best_conf * 100, 2)}%")
|
| 356 |
+
return best_weight, round(best_conf * 100, 2)
|
| 357 |
+
|
| 358 |
+
except Exception as e:
|
| 359 |
+
logging.error(f"Weight extraction failed unexpectedly: {str(e)}")
|
| 360 |
+
return "Not detected", 0.0
|