Spaces:
Sleeping
Sleeping
Update ocr_engine.py
Browse files- ocr_engine.py +44 -371
ocr_engine.py
CHANGED
@@ -2,306 +2,54 @@ import pytesseract
|
|
2 |
import numpy as np
|
3 |
import cv2
|
4 |
import re
|
5 |
-
import logging
|
6 |
-
from datetime import datetime
|
7 |
-
import os
|
8 |
from PIL import Image
|
|
|
9 |
|
10 |
# Set up logging
|
11 |
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
12 |
|
13 |
-
# Directory for debug images
|
14 |
-
DEBUG_DIR = "debug_images"
|
15 |
-
os.makedirs(DEBUG_DIR, exist_ok=True)
|
16 |
-
|
17 |
-
def save_debug_image(img, filename_suffix, prefix=""):
|
18 |
-
"""Save image to debug directory with timestamp."""
|
19 |
-
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S_%f")
|
20 |
-
filename = os.path.join(DEBUG_DIR, f"{prefix}{timestamp}_{filename_suffix}.png")
|
21 |
-
if isinstance(img, Image.Image):
|
22 |
-
img.save(filename)
|
23 |
-
elif len(img.shape) == 3:
|
24 |
-
cv2.imwrite(filename, cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
|
25 |
-
else:
|
26 |
-
cv2.imwrite(filename, img)
|
27 |
-
logging.info(f"Saved debug image: {filename}")
|
28 |
-
|
29 |
-
def estimate_brightness(img):
|
30 |
-
"""Estimate image brightness."""
|
31 |
-
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
|
32 |
-
return np.mean(gray)
|
33 |
-
|
34 |
def preprocess_image(img):
|
35 |
-
"""Preprocess image
|
36 |
-
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
|
37 |
-
brightness = estimate_brightness(img)
|
38 |
-
|
39 |
-
# Apply mild CLAHE for contrast
|
40 |
-
clahe_clip = 8.0 if brightness < 90 else 4.0
|
41 |
-
clahe = cv2.createCLAHE(clipLimit=clahe_clip, tileGridSize=(8, 8))
|
42 |
-
enhanced = clahe.apply(gray)
|
43 |
-
save_debug_image(enhanced, "01_preprocess_clahe")
|
44 |
-
|
45 |
-
# Light blur to reduce noise
|
46 |
-
blurred = cv2.GaussianBlur(enhanced, (5, 5), 0)
|
47 |
-
save_debug_image(blurred, "02_preprocess_blur")
|
48 |
-
|
49 |
-
# Dynamic thresholding with larger block size for small displays
|
50 |
-
block_size = max(7, min(31, int(img.shape[0] / 20) * 2 + 1))
|
51 |
-
thresh = cv2.adaptiveThreshold(
|
52 |
-
blurred, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
|
53 |
-
cv2.THRESH_BINARY_INV, block_size, 3
|
54 |
-
)
|
55 |
-
|
56 |
-
# Minimal morphological operations
|
57 |
-
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
|
58 |
-
thresh = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=1)
|
59 |
-
save_debug_image(thresh, "03_preprocess_morph")
|
60 |
-
return thresh, enhanced
|
61 |
-
|
62 |
-
def correct_rotation(img):
|
63 |
-
"""Correct image rotation using edge detection."""
|
64 |
try:
|
|
|
|
|
|
|
|
|
|
|
65 |
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
|
66 |
-
edges = cv2.Canny(gray, 30, 100, apertureSize=3)
|
67 |
-
lines = cv2.HoughLinesP(edges, 1, np.pi / 180, threshold=25, minLineLength=15, maxLineGap=10)
|
68 |
-
if lines is not None:
|
69 |
-
angles = [np.arctan2(line[0][3] - line[0][1], line[0][2] - line[0][0]) * 180 / np.pi for line in lines]
|
70 |
-
angle = np.median(angles)
|
71 |
-
if abs(angle) > 0.3:
|
72 |
-
h, w = img.shape[:2]
|
73 |
-
center = (w // 2, h // 2)
|
74 |
-
M = cv2.getRotationMatrix2D(center, angle, 1.0)
|
75 |
-
img = cv2.warpAffine(img, M, (w, h))
|
76 |
-
save_debug_image(img, "00_rotated_image")
|
77 |
-
logging.info(f"Applied rotation: {angle:.2f} degrees")
|
78 |
-
return img
|
79 |
-
except Exception as e:
|
80 |
-
logging.error(f"Rotation correction failed: {str(e)}")
|
81 |
-
return img
|
82 |
-
|
83 |
-
def detect_roi(img):
|
84 |
-
"""Detect region of interest with broader contour analysis."""
|
85 |
-
try:
|
86 |
-
save_debug_image(img, "04_original")
|
87 |
-
thresh, enhanced = preprocess_image(img)
|
88 |
-
brightness_map = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
|
89 |
-
block_sizes = [max(7, min(31, int(img.shape[0] / s) * 2 + 1)) for s in [5, 10, 20]]
|
90 |
-
valid_contours = []
|
91 |
-
img_area = img.shape[0] * img.shape[1]
|
92 |
|
93 |
-
for
|
94 |
-
|
95 |
-
enhanced, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
|
96 |
-
cv2.THRESH_BINARY_INV, block_size, 3
|
97 |
-
)
|
98 |
-
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
|
99 |
-
temp_thresh = cv2.morphologyEx(temp_thresh, cv2.MORPH_CLOSE, kernel, iterations=2)
|
100 |
-
save_debug_image(temp_thresh, f"05_roi_threshold_block{block_size}")
|
101 |
-
contours, _ = cv2.findContours(temp_thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
|
102 |
-
|
103 |
-
for c in contours:
|
104 |
-
area = cv2.contourArea(c)
|
105 |
-
x, y, w, h = cv2.boundingRect(c)
|
106 |
-
roi_brightness = np.mean(brightness_map[y:y+h, x:x+w])
|
107 |
-
aspect_ratio = w / h
|
108 |
-
if (50 < area < (img_area * 0.95) and
|
109 |
-
0.05 <= aspect_ratio <= 20.0 and w > 20 and h > 8 and roi_brightness > 15):
|
110 |
-
valid_contours.append((c, area * roi_brightness))
|
111 |
-
logging.debug(f"Contour (block {block_size}): Area={area}, Aspect={aspect_ratio:.2f}, Brightness={roi_brightness:.2f}")
|
112 |
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
x, y = max(0, x - padding), max(0, y - padding)
|
118 |
-
w, h = min(w + 2 * padding, img.shape[1] - x), min(h + 2 * padding, img.shape[0] - y)
|
119 |
-
roi_img = img[y:y+h, x:x+w]
|
120 |
-
save_debug_image(roi_img, "06_detected_roi")
|
121 |
-
logging.info(f"Detected ROI: ({x}, {y}, {w}, {h})")
|
122 |
-
return roi_img, (x, y, w, h)
|
123 |
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
h, w = digit_img.shape
|
136 |
-
if h < 5 or w < 2:
|
137 |
-
logging.debug("Digit image too small for template matching.")
|
138 |
-
return None
|
139 |
-
|
140 |
-
# Expanded digit templates for seven-segment display variations
|
141 |
-
digit_templates = {
|
142 |
-
'0': [
|
143 |
-
np.array([[1, 1, 1, 1, 1],
|
144 |
-
[1, 0, 0, 0, 1],
|
145 |
-
[1, 0, 0, 0, 1],
|
146 |
-
[1, 0, 0, 0, 1],
|
147 |
-
[1, 1, 1, 1, 1]], dtype=np.float32),
|
148 |
-
np.array([[1, 1, 1, 1],
|
149 |
-
[1, 0, 0, 1],
|
150 |
-
[1, 0, 0, 1],
|
151 |
-
[1, 0, 0, 1],
|
152 |
-
[1, 1, 1, 1]], dtype=np.float32)
|
153 |
-
],
|
154 |
-
'1': [
|
155 |
-
np.array([[0, 0, 1, 0, 0],
|
156 |
-
[0, 0, 1, 0, 0],
|
157 |
-
[0, 0, 1, 0, 0],
|
158 |
-
[0, 0, 1, 0, 0],
|
159 |
-
[0, 0, 1, 0, 0]], dtype=np.float32),
|
160 |
-
np.array([[0, 1, 0],
|
161 |
-
[0, 1, 0],
|
162 |
-
[0, 1, 0],
|
163 |
-
[0, 1, 0],
|
164 |
-
[0, 1, 0]], dtype=np.float32)
|
165 |
-
],
|
166 |
-
'2': [
|
167 |
-
np.array([[1, 1, 1, 1, 1],
|
168 |
-
[0, 0, 0, 1, 1],
|
169 |
-
[1, 1, 1, 1, 1],
|
170 |
-
[1, 1, 0, 0, 0],
|
171 |
-
[1, 1, 1, 1, 1]], dtype=np.float32),
|
172 |
-
np.array([[1, 1, 1, 1],
|
173 |
-
[0, 0, 1, 1],
|
174 |
-
[1, 1, 1, 1],
|
175 |
-
[1, 1, 0, 0],
|
176 |
-
[1, 1, 1, 1]], dtype=np.float32)
|
177 |
-
],
|
178 |
-
'3': [
|
179 |
-
np.array([[1, 1, 1, 1, 1],
|
180 |
-
[0, 0, 0, 1, 1],
|
181 |
-
[1, 1, 1, 1, 1],
|
182 |
-
[0, 0, 0, 1, 1],
|
183 |
-
[1, 1, 1, 1, 1]], dtype=np.float32),
|
184 |
-
np.array([[1, 1, 1, 1],
|
185 |
-
[0, 0, 1, 1],
|
186 |
-
[1, 1, 1, 1],
|
187 |
-
[0, 0, 1, 1],
|
188 |
-
[1, 1, 1, 1]], dtype=np.float32)
|
189 |
-
],
|
190 |
-
'4': [
|
191 |
-
np.array([[1, 1, 0, 0, 1],
|
192 |
-
[1, 1, 0, 0, 1],
|
193 |
-
[1, 1, 1, 1, 1],
|
194 |
-
[0, 0, 0, 0, 1],
|
195 |
-
[0, 0, 0, 0, 1]], dtype=np.float32),
|
196 |
-
np.array([[1, 0, 0, 1],
|
197 |
-
[1, 0, 0, 1],
|
198 |
-
[1, 1, 1, 1],
|
199 |
-
[0, 0, 0, 1],
|
200 |
-
[0, 0, 0, 1]], dtype=np.float32)
|
201 |
-
],
|
202 |
-
'5': [
|
203 |
-
np.array([[1, 1, 1, 1, 1],
|
204 |
-
[1, 1, 0, 0, 0],
|
205 |
-
[1, 1, 1, 1, 1],
|
206 |
-
[0, 0, 0, 1, 1],
|
207 |
-
[1, 1, 1, 1, 1]], dtype=np.float32),
|
208 |
-
np.array([[1, 1, 1, 1],
|
209 |
-
[1, 1, 0, 0],
|
210 |
-
[1, 1, 1, 1],
|
211 |
-
[0, 0, 1, 1],
|
212 |
-
[1, 1, 1, 1]], dtype=np.float32)
|
213 |
-
],
|
214 |
-
'6': [
|
215 |
-
np.array([[1, 1, 1, 1, 1],
|
216 |
-
[1, 1, 0, 0, 0],
|
217 |
-
[1, 1, 1, 1, 1],
|
218 |
-
[1, 0, 0, 1, 1],
|
219 |
-
[1, 1, 1, 1, 1]], dtype=np.float32),
|
220 |
-
np.array([[1, 1, 1, 1],
|
221 |
-
[1, 1, 0, 0],
|
222 |
-
[1, 1, 1, 1],
|
223 |
-
[1, 0, 1, 1],
|
224 |
-
[1, 1, 1, 1]], dtype=np.float32)
|
225 |
-
],
|
226 |
-
'7': [
|
227 |
-
np.array([[1, 1, 1, 1, 1],
|
228 |
-
[0, 0, 0, 0, 1],
|
229 |
-
[0, 0, 0, 0, 1],
|
230 |
-
[0, 0, 0, 0, 1],
|
231 |
-
[0, 0, 0, 0, 1]], dtype=np.float32),
|
232 |
-
np.array([[1, 1, 1, 1],
|
233 |
-
[0, 0, 0, 1],
|
234 |
-
[0, 0, 0, 1],
|
235 |
-
[0, 0, 0, 1],
|
236 |
-
[0, 0, 0, 1]], dtype=np.float32)
|
237 |
-
],
|
238 |
-
'8': [
|
239 |
-
np.array([[1, 1, 1, 1, 1],
|
240 |
-
[1, 0, 0, 0, 1],
|
241 |
-
[1, 1, 1, 1, 1],
|
242 |
-
[1, 0, 0, 0, 1],
|
243 |
-
[1, 1, 1, 1, 1]], dtype=np.float32),
|
244 |
-
np.array([[1, 1, 1, 1],
|
245 |
-
[1, 0, 0, 1],
|
246 |
-
[1, 1, 1, 1],
|
247 |
-
[1, 0, 0, 1],
|
248 |
-
[1, 1, 1, 1]], dtype=np.float32)
|
249 |
-
],
|
250 |
-
'9': [
|
251 |
-
np.array([[1, 1, 1, 1, 1],
|
252 |
-
[1, 0, 0, 0, 1],
|
253 |
-
[1, 1, 1, 1, 1],
|
254 |
-
[0, 0, 0, 1, 1],
|
255 |
-
[1, 1, 1, 1, 1]], dtype=np.float32),
|
256 |
-
np.array([[1, 1, 1, 1],
|
257 |
-
[1, 0, 0, 1],
|
258 |
-
[1, 1, 1, 1],
|
259 |
-
[0, 0, 1, 1],
|
260 |
-
[1, 1, 1, 1]], dtype=np.float32)
|
261 |
-
],
|
262 |
-
'.': [
|
263 |
-
np.array([[0, 0, 0],
|
264 |
-
[0, 1, 0],
|
265 |
-
[0, 0, 0]], dtype=np.float32),
|
266 |
-
np.array([[0, 0],
|
267 |
-
[1, 0],
|
268 |
-
[0, 0]], dtype=np.float32)
|
269 |
-
]
|
270 |
-
}
|
271 |
-
|
272 |
-
# Try multiple sizes for digit image
|
273 |
-
sizes = [(5, 5), (4, 4), (3, 3)] if h > w else [(3, 3), (2, 2)]
|
274 |
-
best_match, best_score = None, -1
|
275 |
-
for size in sizes:
|
276 |
-
digit_img_resized = cv2.resize(digit_img, size, interpolation=cv2.INTER_AREA)
|
277 |
-
digit_img_resized = (digit_img_resized > 100).astype(np.float32) # Binarize
|
278 |
-
|
279 |
-
for digit, templates in digit_templates.items():
|
280 |
-
for template in templates:
|
281 |
-
if digit == '.' and size[0] > 3:
|
282 |
-
continue
|
283 |
-
if digit != '.' and size[0] <= 3:
|
284 |
-
continue
|
285 |
-
if template.shape[0] != size[0] or template.shape[1] != size[1]:
|
286 |
-
continue
|
287 |
-
result = cv2.matchTemplate(digit_img_resized, template, cv2.TM_CCOEFF_NORMED)
|
288 |
-
_, max_val, _, _ = cv2.minMaxLoc(result)
|
289 |
-
if max_val > 0.55 and max_val > best_score: # Further lowered threshold
|
290 |
-
best_score = max_val
|
291 |
-
best_match = digit
|
292 |
-
logging.debug(f"Template match: {best_match}, Score: {best_score:.2f}")
|
293 |
-
return best_match if best_score > 0.55 else None
|
294 |
except Exception as e:
|
295 |
-
logging.error(f"
|
296 |
-
return
|
297 |
|
298 |
-
def
|
299 |
-
"""
|
300 |
try:
|
301 |
-
|
302 |
-
|
303 |
-
|
304 |
-
|
|
|
|
|
305 |
|
306 |
# Try multiple Tesseract configurations
|
307 |
configs = [
|
@@ -309,8 +57,10 @@ def perform_ocr(img, roi_bbox):
|
|
309 |
r'--oem 3 --psm 6 -c tessedit_char_whitelist=0123456789.' # Block of text
|
310 |
]
|
311 |
for config in configs:
|
312 |
-
text = pytesseract.image_to_string(
|
313 |
logging.info(f"Tesseract raw output (config {config}): {text}")
|
|
|
|
|
314 |
text = re.sub(r"[^\d\.]", "", text)
|
315 |
if text.count('.') > 1:
|
316 |
text = text.replace('.', '', text.count('.') - 1)
|
@@ -318,91 +68,14 @@ def perform_ocr(img, roi_bbox):
|
|
318 |
if text and re.fullmatch(r"^\d*\.?\d*$", text):
|
319 |
text = text.lstrip('0') or '0'
|
320 |
confidence = 95.0 if len(text.replace('.', '')) >= 3 else 90.0
|
321 |
-
|
322 |
-
|
323 |
-
|
324 |
-
|
325 |
-
|
326 |
-
|
327 |
-
|
328 |
-
for c in contours:
|
329 |
-
x, y, w, h = cv2.boundingRect(c)
|
330 |
-
if w > 4 and h > 5 and 0.03 <= w/h <= 4.0:
|
331 |
-
digits_info.append((x, x+w, y, y+h))
|
332 |
-
|
333 |
-
if digits_info:
|
334 |
-
digits_info.sort(key=lambda x: x[0])
|
335 |
-
recognized_text = ""
|
336 |
-
prev_x_max = -float('inf')
|
337 |
-
for idx, (x_min, x_max, y_min, y_max) in enumerate(digits_info):
|
338 |
-
x_min, y_min = max(0, x_min), max(0, y_min)
|
339 |
-
x_max, y_max = min(thresh.shape[1], x_max), min(thresh.shape[0], y_max)
|
340 |
-
if x_max <= x_min or y_max <= y_min:
|
341 |
-
continue
|
342 |
-
digit_crop = thresh[y_min:y_max, x_min:x_max]
|
343 |
-
save_debug_image(digit_crop, f"08_digit_crop_{idx}")
|
344 |
-
digit = detect_digit_template(digit_crop, brightness)
|
345 |
-
if digit:
|
346 |
-
recognized_text += digit
|
347 |
-
elif x_min - prev_x_max < 10 and prev_x_max != -float('inf'):
|
348 |
-
recognized_text += '.'
|
349 |
-
prev_x_max = x_max
|
350 |
-
|
351 |
-
text = re.sub(r"[^\d\.]", "", recognized_text)
|
352 |
-
if text.count('.') > 1:
|
353 |
-
text = text.replace('.', '', text.count('.') - 1)
|
354 |
-
text = text.strip('.')
|
355 |
-
if text and re.fullmatch(r"^\d*\.?\d*$", text):
|
356 |
-
text = text.lstrip('0') or '0'
|
357 |
-
confidence = 90.0 if len(text.replace('.', '')) >= 3 else 85.0
|
358 |
-
logging.info(f"Validated template text: {text}, Confidence: {confidence:.2f}%")
|
359 |
-
return text, confidence
|
360 |
|
361 |
-
logging.info("No valid digits detected.")
|
362 |
-
return None, 0.0
|
363 |
-
except Exception as e:
|
364 |
-
logging.error(f"OCR failed: {str(e)}")
|
365 |
-
return None, 0.0
|
366 |
-
|
367 |
-
def extract_weight_from_image(pil_img):
|
368 |
-
"""Extract weight from any digital scale image."""
|
369 |
-
try:
|
370 |
-
img = np.array(pil_img)
|
371 |
-
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
|
372 |
-
save_debug_image(img, "00_input_image")
|
373 |
-
img = correct_rotation(img)
|
374 |
-
brightness = estimate_brightness(img)
|
375 |
-
conf_threshold = 0.65 if brightness > 70 else 0.45
|
376 |
-
|
377 |
-
# Try ROI-based detection
|
378 |
-
roi_img, roi_bbox = detect_roi(img)
|
379 |
-
if roi_bbox:
|
380 |
-
conf_threshold *= 1.15 if (roi_bbox[2] * roi_bbox[3]) > (img.shape[0] * img.shape[1] * 0.05) else 1.0
|
381 |
-
|
382 |
-
result, confidence = perform_ocr(roi_img, roi_bbox)
|
383 |
-
if result and confidence >= conf_threshold * 100:
|
384 |
-
try:
|
385 |
-
weight = float(result)
|
386 |
-
if 0.001 <= weight <= 5000:
|
387 |
-
logging.info(f"Detected weight: {result} kg, Confidence: {confidence:.2f}%")
|
388 |
-
return result, confidence
|
389 |
-
logging.warning(f"Weight {result} out of range.")
|
390 |
-
except ValueError:
|
391 |
-
logging.warning(f"Invalid weight format: {result}")
|
392 |
-
|
393 |
-
# Full image fallback
|
394 |
-
logging.info("Primary OCR failed, using full image fallback.")
|
395 |
-
result, confidence = perform_ocr(img, None)
|
396 |
-
if result and confidence >= conf_threshold * 0.85 * 100:
|
397 |
-
try:
|
398 |
-
weight = float(result)
|
399 |
-
if 0.001 <= weight <= 5000:
|
400 |
-
logging.info(f"Full image weight: {result} kg, Confidence: {confidence:.2f}%")
|
401 |
-
return result, confidence
|
402 |
-
logging.warning(f"Full image weight {result} out of range.")
|
403 |
-
except ValueError:
|
404 |
-
logging.warning(f"Invalid full image weight format: {result}")
|
405 |
-
|
406 |
logging.info("No valid weight detected.")
|
407 |
return "Not detected", 0.0
|
408 |
except Exception as e:
|
|
|
2 |
import numpy as np
|
3 |
import cv2
|
4 |
import re
|
|
|
|
|
|
|
5 |
from PIL import Image
|
6 |
+
import logging
|
7 |
|
8 |
# Set up logging
|
9 |
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
def preprocess_image(img):
|
12 |
+
"""Preprocess image for robust OCR."""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
try:
|
14 |
+
# Convert to OpenCV format
|
15 |
+
img = np.array(img)
|
16 |
+
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
|
17 |
+
|
18 |
+
# Convert to grayscale
|
19 |
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
+
# Estimate brightness for adaptive processing
|
22 |
+
brightness = np.mean(gray)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
|
24 |
+
# Apply CLAHE for contrast enhancement
|
25 |
+
clahe_clip = 4.0 if brightness < 100 else 2.0
|
26 |
+
clahe = cv2.createCLAHE(clipLimit=clahe_clip, tileGridSize=(8, 8))
|
27 |
+
enhanced = clahe.apply(gray)
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
|
29 |
+
# Apply adaptive thresholding
|
30 |
+
block_size = max(11, min(31, int(img.shape[0] / 20) * 2 + 1))
|
31 |
+
thresh = cv2.adaptiveThreshold(
|
32 |
+
enhanced, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, block_size, 2
|
33 |
+
)
|
34 |
+
|
35 |
+
# Noise reduction
|
36 |
+
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
|
37 |
+
thresh = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=1)
|
38 |
+
|
39 |
+
return thresh
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
except Exception as e:
|
41 |
+
logging.error(f"Preprocessing failed: {str(e)}")
|
42 |
+
return img
|
43 |
|
44 |
+
def extract_weight_from_image(pil_img):
|
45 |
+
"""Extract weight from any digital scale image."""
|
46 |
try:
|
47 |
+
# Convert PIL image to OpenCV
|
48 |
+
img = np.array(pil_img)
|
49 |
+
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
|
50 |
+
|
51 |
+
# Preprocess image
|
52 |
+
thresh = preprocess_image(img)
|
53 |
|
54 |
# Try multiple Tesseract configurations
|
55 |
configs = [
|
|
|
57 |
r'--oem 3 --psm 6 -c tessedit_char_whitelist=0123456789.' # Block of text
|
58 |
]
|
59 |
for config in configs:
|
60 |
+
text = pytesseract.image_to_string(thresh, config=config)
|
61 |
logging.info(f"Tesseract raw output (config {config}): {text}")
|
62 |
+
|
63 |
+
# Clean and validate text
|
64 |
text = re.sub(r"[^\d\.]", "", text)
|
65 |
if text.count('.') > 1:
|
66 |
text = text.replace('.', '', text.count('.') - 1)
|
|
|
68 |
if text and re.fullmatch(r"^\d*\.?\d*$", text):
|
69 |
text = text.lstrip('0') or '0'
|
70 |
confidence = 95.0 if len(text.replace('.', '')) >= 3 else 90.0
|
71 |
+
try:
|
72 |
+
weight = float(text)
|
73 |
+
if 0.001 <= weight <= 5000:
|
74 |
+
logging.info(f"Detected weight: {text} kg, Confidence: {confidence:.2f}%")
|
75 |
+
return text, confidence
|
76 |
+
except ValueError:
|
77 |
+
logging.warning(f"Invalid weight format: {text}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
78 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
79 |
logging.info("No valid weight detected.")
|
80 |
return "Not detected", 0.0
|
81 |
except Exception as e:
|