-
Notifications
You must be signed in to change notification settings - Fork 2
/
tesseract.py
574 lines (534 loc) · 22 KB
/
tesseract.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
from os import path
import cv2
from tesserocr import PyTessBaseAPI, RIL, iterate_level
import numpy as np
from PIL import Image
from defaults import FieldType
from storage import fetch_data
from text_detection_target import TextDetectionTarget, TextDetectionTargetWithResult
import re
from PySide6.QtCore import QRectF
from threading import Lock
from sc_logging import logger
def autocrop(image_in):
image = image_in.copy()
# check if this image is black-on-white or white-on-black by looking at the first few pixels
if np.sum(image[0:5, 0:5]) > 0:
# black-on-white
# invert the image
image = 255 - image
# find the first row that has a pixel
first_row = 0
for row in range(image.shape[0]):
if np.sum(image[row, :]) > 0:
first_row = row
break
# find the last row that has a pixel
last_row = image.shape[0] - 1
for row in range(image.shape[0] - 1, -1, -1):
if np.sum(image[row, :]) > 0:
last_row = row
break
# find the first column that has a pixel
first_col = 0
for col in range(image.shape[1]):
if np.sum(image[:, col]) > 0:
first_col = col
break
# find the last column that has a pixel
last_col = image.shape[1] - 1
for col in range(image.shape[1] - 1, -1, -1):
if np.sum(image[:, col]) > 0:
last_col = col
break
# leave a 10 pixel border on each side
first_row = max(0, first_row - 10)
last_row = min(image.shape[0] - 1, last_row + 10)
first_col = max(0, first_col - 10)
last_col = min(image.shape[1] - 1, last_col + 10)
return image_in[first_row:last_row, first_col:last_col], (
first_row,
last_row,
first_col,
last_col,
)
def add_ordinal_indicator(text):
if text == "":
return ""
if text.endswith("1") and text != "11":
return text + "st"
elif text.endswith("2") and text != "12":
return text + "nd"
elif text.endswith("3") and text != "13":
return text + "rd"
else:
return text + "th"
def is_valid_regex(pattern):
try:
re.compile(pattern)
return True
except re.error:
return False
class TextDetectionResult:
def __init__(self, text, state, rect=None, extra=None):
self.text = text
self.state = state
self.rect = rect
self.extra = extra
class TextDetector:
# model name enum: daktronics=0, scoreboard_general=1
class OcrModelIndex:
DAKTRONICS = 0
SCOREBOARD_GENERAL = 1
GENERAL_ENGLISH = 2
SCOREBOARD_GENERAL_LARGE = 3
class BinarizationMethod:
GLOBAL = 0
NO_BINARIZATION = 1
LOCAL = 2
ADAPTIVE = 3
def __init__(self):
self.api_lock = Lock()
self.api = None
if (
fetch_data(
"scoresight.json",
"ocr_model",
TextDetector.OcrModelIndex.SCOREBOARD_GENERAL,
)
== TextDetector.OcrModelIndex.SCOREBOARD_GENERAL
):
self.setOcrModel(TextDetector.OcrModelIndex.SCOREBOARD_GENERAL)
else:
self.setOcrModel(TextDetector.OcrModelIndex.DAKTRONICS)
def setOcrModel(self, ocrModelIndex):
ocr_model = None
if ocrModelIndex == self.OcrModelIndex.DAKTRONICS:
ocr_model = "daktronics"
if ocrModelIndex == self.OcrModelIndex.SCOREBOARD_GENERAL:
ocr_model = "scoreboard_general"
if ocrModelIndex == self.OcrModelIndex.GENERAL_ENGLISH:
ocr_model = "eng"
if ocrModelIndex == self.OcrModelIndex.SCOREBOARD_GENERAL_LARGE:
ocr_model = "scoreboard_general_large"
if ocr_model is None:
return
with self.api_lock:
if self.api is not None:
self.api.End()
self.api = None
self.api = PyTessBaseAPI(
path=path.abspath(
path.join(path.dirname(__file__), "tesseract/tessdata")
),
lang=ocr_model,
)
# single word PSM
self.api.SetPageSegMode(8)
self.api.SetVariable("load_system_dawg", "F")
self.api.SetVariable("load_freq_dawg", "F")
def detect_text(self, image):
if image is None:
return ""
if not isinstance(image, np.ndarray):
return ""
# check the image has rows and columns
if len(image.shape) < 2 or image.shape[0] < 1 or image.shape[1] < 1:
return ""
pilimage = Image.fromarray(image)
text = ""
with self.api_lock:
self.api.SetImage(pilimage)
text = self.api.GetUTF8Text()
return text.strip()
def detect_multi_text(
self, binary, gray, rects: list[TextDetectionTarget]
) -> list[TextDetectionResult]:
if binary is None:
return []
if not isinstance(binary, np.ndarray):
return []
# check the image has rows and columns
if len(binary.shape) < 2 or binary.shape[0] < 1 or binary.shape[1] < 1:
return []
texts = []
for rect in rects:
effectiveRect = None
scale_x = 1.0
scale_y = 1.0
if (
rect is None
or rect.x() < 0
or rect.y() < 0
or rect.width() < 1
or rect.height() < 1
):
texts.append(
TextDetectionResult(
"", TextDetectionTargetWithResult.ResultState.Empty, None
)
)
continue
if rect.x() >= binary.shape[1]:
# move the rect inside the image
rect.setX(binary.shape[1] - rect.width())
if rect.y() >= binary.shape[0]:
# move the rect inside the image
rect.setY(binary.shape[0] - rect.height())
if rect.x() + rect.width() > binary.shape[1]:
rect.setWidth(binary.shape[1] - rect.x())
if rect.y() + rect.height() > binary.shape[0]:
rect.setHeight(binary.shape[0] - rect.y())
if (
rect.settings is not None
and "binarization_method" in rect.settings
and rect.settings["binarization_method"]
!= TextDetector.BinarizationMethod.GLOBAL
):
if (
rect.settings["binarization_method"]
== TextDetector.BinarizationMethod.NO_BINARIZATION
):
# no binarization
imagecrop = gray[
int(rect.y()) : int(rect.y() + rect.height()),
int(rect.x()) : int(rect.x() + rect.width()),
]
elif (
rect.settings["binarization_method"]
== TextDetector.BinarizationMethod.LOCAL
):
# local binarization using Otsu's method
_, imagecrop = cv2.threshold(
gray[
int(rect.y()) : int(rect.y() + rect.height()),
int(rect.x()) : int(rect.x() + rect.width()),
],
0,
255,
cv2.THRESH_BINARY + cv2.THRESH_OTSU,
)
elif (
rect.settings["binarization_method"]
== TextDetector.BinarizationMethod.ADAPTIVE
):
# apply adaptive binarization
imagecrop = cv2.adaptiveThreshold(
gray[
int(rect.y()) : int(rect.y() + rect.height()),
int(rect.x()) : int(rect.x() + rect.width()),
],
255,
cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY,
# use a fraction of the patch area
max(int(rect.width() * rect.height() * 0.01), 3) | 1,
2,
)
# update the binary image for visualisation in the binary mode
binary[
int(rect.y()) : int(rect.y() + rect.height()),
int(rect.x()) : int(rect.x() + rect.width()),
] = imagecrop
else:
imagecrop = binary[
int(rect.y()) : int(rect.y() + rect.height()),
int(rect.x()) : int(rect.x() + rect.width()),
]
if (
rect.settings is not None
and "cleanup_thresh" in rect.settings
and rect.settings["cleanup_thresh"] > 0
):
# cleanup image from small components: find contours and remove small ones
contours, _ = cv2.findContours(
imagecrop, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
)
# cleanup_thresh is [0, 1.0], convert to [0, 0.05]
cleanup_thresh = rect.settings["cleanup_thresh"] * 0.05
img_area_thresh = (
imagecrop.shape[0] * imagecrop.shape[1] * cleanup_thresh
)
for contour in contours:
if cv2.contourArea(contour) < img_area_thresh:
cv2.drawContours(imagecrop, [contour], 0, 0, -1)
if (
rect.settings is not None
and "vscale" in rect.settings
and rect.settings["vscale"] != 10
):
# vertical scale the image
# the vscale input is in the range [1, 10] where 10 is the default (1:1)
# scale the image in the y direction about the center
rows, cols = imagecrop.shape
# calculate the target height
target_height = int(rows * (rect.settings["vscale"] / 10.0))
scaled = cv2.resize(
imagecrop, (cols, target_height), 0, 0, cv2.INTER_AREA
)
# add padding to the top and bottom
pad_top = (rows - target_height) // 2
pad_bottom = rows - target_height - pad_top
scaled = cv2.copyMakeBorder(
scaled, pad_top, pad_bottom, 0, 0, cv2.BORDER_REPLICATE
)
# make sure the image is the same size as the original
scaled = scaled[:rows, :]
# copy back into imagecrop and binary display
binary[
int(rect.y()) : int(rect.y() + rect.height()),
int(rect.x()) : int(rect.x() + rect.width()),
] = scaled
imagecrop = scaled
if (
rect.settings is not None
and "skew" in rect.settings
and rect.settings["skew"] != 0
):
# skew the image in the x direction about the center
rows, cols = imagecrop.shape
# identity 2x2 matrix
M = np.float32([[1, 0, 0], [0, 1, 0]])
# add skew factor to matrix
M[0, 1] = rect.settings["skew"] / 40.0
try:
skewed = cv2.warpAffine(imagecrop, M, (cols, rows))
binary[
int(rect.y()) : int(rect.y() + rect.height()),
int(rect.x()) : int(rect.x() + rect.width()),
] = skewed
imagecrop = skewed
except:
pass
if (
rect.settings is not None
and "dilate" in rect.settings
and rect.settings["dilate"] > 0
and imagecrop.shape[0] > 0
and imagecrop.shape[1] > 0
):
# dilate the image
kernel = np.ones((3, 3), np.uint8)
dilated = cv2.dilate(
imagecrop.copy(),
kernel,
iterations=int(rect.settings["dilate"]),
)
# copy back into image crop
binary[
int(rect.y()) : int(rect.y() + rect.height()),
int(rect.x()) : int(rect.x() + rect.width()),
] = dilated
if (
rect.settings is not None
and "invert_patch" in rect.settings
and rect.settings["invert_patch"]
):
# invert the image
imagecrop = 255 - imagecrop
if (
rect.settings is not None
and "skip_similar_image" in rect.settings
and rect.settings["skip_similar_image"]
):
# compare the image with the last image
if (
rect.last_image is not None
and rect.last_image.shape == imagecrop.shape
):
# check if the difference is less than 5%
diff = cv2.absdiff(rect.last_image, imagecrop)
diff = diff.astype(np.float32)
diff = diff / 255.0
diff = diff.sum() / (imagecrop.shape[0] * imagecrop.shape[1])
if diff < 0.05:
# skip this image
texts.append(
TextDetectionResult(
"SIM",
TextDetectionTargetWithResult.ResultState.FailedFilter,
effectiveRect,
)
)
continue
rect.last_image = imagecrop.copy()
if (
rect.settings is not None
and "autocrop" in rect.settings
and rect.settings["autocrop"]
):
# auto crop the binary image around the text
imagecrop, (first_row, last_row, first_col, last_col) = autocrop(
imagecrop
)
effectiveRect = QRectF(
first_col,
first_row,
last_col - first_col,
last_row - first_row,
)
# check if image is size 0
if imagecrop.shape[0] == 0 or imagecrop.shape[1] == 0:
texts.append(
TextDetectionResult(
"",
TextDetectionTargetWithResult.ResultState.Empty,
effectiveRect,
)
)
continue
if (
rect.settings is not None
and "rescale_patch" in rect.settings
and rect.settings["rescale_patch"]
):
# rescale the image to 35 pixels height
scale_x = 35 / imagecrop.shape[0]
scale_y = scale_x
if (
rect.settings is not None
and "normalize_wh_ratio" in rect.settings
and rect.settings["normalize_wh_ratio"]
and "median_wh_ratio" in rect.settings
and rect.settings["median_wh_ratio"] > 0
):
# rescale the image in x or in y such that the width-to-height ratio is 0.5
scale_x *= 0.5 / rect.settings["median_wh_ratio"]
if scale_x != 1.0 or scale_y != 1.0:
imagecrop = cv2.resize(
imagecrop,
None,
fx=scale_x,
fy=scale_y,
interpolation=cv2.INTER_AREA,
)
# if dot detector count the blobs in the patch
if (
rect.settings is not None
and "dot_detector" in rect.settings
and rect.settings["dot_detector"]
):
# find the contours
contours, _ = cv2.findContours(
imagecrop, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
)
# count the number of contours
count = 0
for contour in contours:
if cv2.contourArea(contour) > 5:
count += 1
texts.append(
TextDetectionResult(
str(count),
TextDetectionTargetWithResult.ResultState.Success,
effectiveRect,
)
)
continue
try:
pilimage = Image.fromarray(imagecrop)
with self.api_lock:
self.api.SetImage(pilimage)
except:
texts.append(
TextDetectionResult(
"", TextDetectionTargetWithResult.ResultState.Empty, None
)
)
continue
if rect.settings["type"] == FieldType.NUMBER:
with self.api_lock:
self.api.SetVariable("tessedit_char_whitelist", "0123456789")
elif rect.settings["type"] == FieldType.TIME:
with self.api_lock:
self.api.SetVariable("tessedit_char_whitelist", "0123456789:.")
else: # general
with self.api_lock:
self.api.SetVariable(
"tessedit_char_whitelist",
"0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ .,;:!?-_()[]{}<>@#$%^&*+=|\\~`\"'",
)
text = ""
extras = {}
with self.api_lock:
text = self.api.GetUTF8Text().strip()
if text != "":
# get the per-character boxes using an iterator with RIL_SYMBOL level
it = self.api.GetIterator()
extras["boxes"] = []
wh_ratios = []
for w in iterate_level(it, RIL.SYMBOL):
char = w.GetUTF8Text(RIL.SYMBOL)
box_tuple = w.BoundingBox(RIL.SYMBOL)
if (
box_tuple is None
or char is None
or char == ""
or len(box_tuple) != 4
):
continue
box = {
"x": box_tuple[0],
"y": box_tuple[1],
"w": box_tuple[2] - box_tuple[0],
"h": box_tuple[3] - box_tuple[1],
}
# box is a dict with x, y, w and h
if scale_x != 1.0 or scale_y != 1.0:
box["x"] = int(box["x"] / scale_x)
box["y"] = int(box["y"] / scale_y)
box["w"] = int(box["w"] / scale_x)
box["h"] = int(box["h"] / scale_y)
if effectiveRect is not None:
box["x"] = int(box["x"] + effectiveRect.x())
box["y"] = int(box["y"] + effectiveRect.y())
extras["boxes"].append(box)
# if char is a "wide character" (like 0,2,3,4,5,6,7,8,9), add the width-to-height ratio
if char in "023456789" and box["h"] > 0:
wh_ratios.append(box["w"] / box["h"])
if (
"normalize_wh_ratio" in rect.settings
and rect.settings["normalize_wh_ratio"]
and "median_wh_ratio" not in rect.settings
and len(wh_ratios) > 0
):
rect.settings["median_wh_ratio"] = np.median(wh_ratios)
textstate = TextDetectionTargetWithResult.ResultState.Success
if rect.settings is not None:
if "format_regex" in rect.settings:
# validate the regex format is valid
if is_valid_regex(rect.settings["format_regex"]):
# check the text matches the regex fully
if not re.fullmatch(rect.settings["format_regex"], text):
textstate = (
TextDetectionTargetWithResult.ResultState.FailedFilter
)
if "conf_thresh" in rect.settings:
with self.api_lock:
meanConf = self.api.MeanTextConf()
if meanConf < rect.settings["conf_thresh"]:
textstate = (
TextDetectionTargetWithResult.ResultState.FailedFilter
)
if "smoothing" in rect.settings:
if rect.settings["smoothing"]:
# apply smoother
text = rect.ocrResultPerCharacterSmoother.get_smoothed_result(
text
)
if text is None:
text = ""
if "remove_leading_zeros" in rect.settings:
if rect.settings["remove_leading_zeros"]:
# remove leading zeros
text = text.lstrip("0")
if text == "":
text = "0"
if "ordinal_indicator" in rect.settings:
if rect.settings["ordinal_indicator"]:
# add ordinal indicator
text = add_ordinal_indicator(text)
if text == "":
textstate = TextDetectionTargetWithResult.ResultState.Empty
texts.append(TextDetectionResult(text, textstate, effectiveRect, extras))
return texts