Spaces:
Running
Running
added queing mechanism
Browse files- extractpuzzle.py +0 -787
- main.py +64 -63
- requirements.txt +0 -2
extractpuzzle.py
DELETED
@@ -1,787 +0,0 @@
|
|
1 |
-
import cv2
|
2 |
-
import numpy as np
|
3 |
-
import math
|
4 |
-
from sklearn.linear_model import LinearRegression
|
5 |
-
import pytesseract
|
6 |
-
import re
|
7 |
-
|
8 |
-
|
9 |
-
pytesseract.pytesseract.tesseract_cmd = "/usr/bin/tesseract"
|
10 |
-
|
11 |
-
def first_preprocessing(image):
|
12 |
-
gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)
|
13 |
-
canny = cv2.Canny(gray,75,25)
|
14 |
-
contours,hierarchies = cv2.findContours(canny,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_NONE)
|
15 |
-
sorted_contours = sorted(contours,key = cv2.contourArea,reverse = True)
|
16 |
-
largest_contour = sorted_contours[0]
|
17 |
-
box = cv2.boundingRect(sorted_contours[0])
|
18 |
-
x = box[0]
|
19 |
-
y = box[1]
|
20 |
-
w = box[2]
|
21 |
-
h = box[3]
|
22 |
-
result = cv2.rectangle(image, (x, y), (x + w, y + h), (255, 255, 255), -1)
|
23 |
-
return result
|
24 |
-
|
25 |
-
def remove_head(image):
|
26 |
-
custom_config = r'--oem 3 --psm 6' # Tesseract OCR configuration
|
27 |
-
detected_text = pytesseract.image_to_string(image, config=custom_config)
|
28 |
-
lines = detected_text.split('\n')
|
29 |
-
|
30 |
-
# Find the first line containing some text
|
31 |
-
line_index = 0
|
32 |
-
for i, line in enumerate(lines):
|
33 |
-
if line.strip() != '':
|
34 |
-
line_index = i
|
35 |
-
break
|
36 |
-
first_newline_idx = detected_text.find('\n')
|
37 |
-
result = cv2.rectangle(image, (0, line_index), (image.shape[1], first_newline_idx), (255,255,255), thickness=cv2.FILLED)
|
38 |
-
return result
|
39 |
-
|
40 |
-
def second_preprocessing(image):
|
41 |
-
gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)
|
42 |
-
canny = cv2.Canny(gray,75,25)
|
43 |
-
contours,hierarchies = cv2.findContours(canny,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_NONE)
|
44 |
-
sorted_contours = sorted(contours,key = cv2.contourArea,reverse = True)
|
45 |
-
largest_contour = sorted_contours[0]
|
46 |
-
box2 = cv2.boundingRect(sorted_contours[0])
|
47 |
-
x = box2[0]
|
48 |
-
y = box2[1]
|
49 |
-
w = box2[2]
|
50 |
-
h = box2[3]
|
51 |
-
result2 = cv2.rectangle(image, (x, y), (x + w, y + h), (255, 255, 255), -1)
|
52 |
-
return result2
|
53 |
-
|
54 |
-
def find_vertical_profile(image):
|
55 |
-
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
|
56 |
-
_, binary = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)
|
57 |
-
vertical_profile = np.sum(binary, axis=0)
|
58 |
-
return vertical_profile
|
59 |
-
|
60 |
-
def detect_steepest_changes(projection_profile, threshold=0.4, start_idx=0, min_valley_width=10, min_search_width=50):
|
61 |
-
differences = np.diff(projection_profile)
|
62 |
-
change_points = np.where(np.abs(differences) > threshold * np.max(np.abs(differences)))[0]
|
63 |
-
left_boundaries = []
|
64 |
-
right_boundaries = []
|
65 |
-
|
66 |
-
for idx in change_points:
|
67 |
-
if idx <= start_idx:
|
68 |
-
continue
|
69 |
-
|
70 |
-
if idx - start_idx >= min_search_width:
|
71 |
-
decreasing_profile = projection_profile[idx:]
|
72 |
-
if np.any(decreasing_profile > 0):
|
73 |
-
right_boundary = idx + np.argmin(decreasing_profile)
|
74 |
-
right_boundaries.append(right_boundary)
|
75 |
-
else:
|
76 |
-
continue
|
77 |
-
valley_start = max(start_idx, idx - min_valley_width)
|
78 |
-
valley_start = valley_start-40
|
79 |
-
valley_end = min(idx + min_valley_width, len(projection_profile) - 1)
|
80 |
-
valley = valley_start + np.argmin(projection_profile[valley_start:valley_end])
|
81 |
-
left_boundaries.append(valley)
|
82 |
-
|
83 |
-
break
|
84 |
-
|
85 |
-
return left_boundaries, right_boundaries
|
86 |
-
|
87 |
-
def crop_text_columns(image, projection_profile, threshold=0.4):
|
88 |
-
start_idx = 0
|
89 |
-
text_columns = []
|
90 |
-
|
91 |
-
while True:
|
92 |
-
left_boundaries, right_boundaries = detect_steepest_changes(projection_profile, threshold, start_idx)
|
93 |
-
if not left_boundaries or not right_boundaries:
|
94 |
-
break
|
95 |
-
left = left_boundaries[0]
|
96 |
-
right = right_boundaries[0]
|
97 |
-
text_column = image[:, left:right]
|
98 |
-
text_columns.append(text_column)
|
99 |
-
|
100 |
-
start_idx = right
|
101 |
-
|
102 |
-
return text_columns
|
103 |
-
|
104 |
-
|
105 |
-
def parse_clues(clue_text):
|
106 |
-
lines = clue_text.split('\n')
|
107 |
-
clues = {}
|
108 |
-
number = None
|
109 |
-
column = 0
|
110 |
-
for line in lines:
|
111 |
-
if "column separation" in line:
|
112 |
-
column += 1
|
113 |
-
continue
|
114 |
-
pattern = r"^(\d+(?:\.\d+)?)\s*(.+)" # Updated pattern to handle decimal point numbers for clues
|
115 |
-
match = re.search(pattern, line)
|
116 |
-
if match:
|
117 |
-
number = float(match.group(1)) # Convert the matched number to float if there is a decimal point
|
118 |
-
if number not in clues:
|
119 |
-
clues[number] = [column,match.group(2).strip()]
|
120 |
-
else:
|
121 |
-
continue
|
122 |
-
elif number is None:
|
123 |
-
continue
|
124 |
-
elif clues[number][0] != column:
|
125 |
-
continue
|
126 |
-
else:
|
127 |
-
clues[number][1] += " " + line.strip() # Append to the previous clue if it's a multiline clue
|
128 |
-
|
129 |
-
return clues
|
130 |
-
|
131 |
-
def parse_crossword_clues(text):
|
132 |
-
# Check if "Down" clues are present
|
133 |
-
match = re.search(r'[dD][oO][wW][nN]\n', text)
|
134 |
-
if match:
|
135 |
-
across_clues, down_clues = re.split(r'[dD][oO][wW][nN]\n', text)
|
136 |
-
else:
|
137 |
-
# If "Down" clues are not present, set down_clues to an empty string
|
138 |
-
across_clues, down_clues = text, ""
|
139 |
-
|
140 |
-
across = parse_clues(across_clues)
|
141 |
-
down = parse_clues(down_clues)
|
142 |
-
|
143 |
-
return across, down
|
144 |
-
|
145 |
-
|
146 |
-
def classify_text(filtered_columns):
|
147 |
-
text = ""
|
148 |
-
custom_config = r'--oem 3 --psm 6'
|
149 |
-
for i, column in enumerate(filtered_columns):
|
150 |
-
column2 = cv2.cvtColor(column, cv2.COLOR_BGR2RGB)
|
151 |
-
scale_factor = 2.0 # You can adjust this value
|
152 |
-
|
153 |
-
# Calculate the new dimensions after scaling
|
154 |
-
new_width = int(column2.shape[1] * scale_factor)
|
155 |
-
new_height = int(column2.shape[0] * scale_factor)
|
156 |
-
|
157 |
-
# Resize the image using OpenCV
|
158 |
-
scaled_image = cv2.resize(column2, (new_width, new_height), interpolation=cv2.INTER_LINEAR)
|
159 |
-
|
160 |
-
# Apply image enhancement techniques
|
161 |
-
denoised_image = cv2.fastNlMeansDenoising(scaled_image, None, h=10, templateWindowSize=7, searchWindowSize=21)
|
162 |
-
enhanced_image = cv2.cvtColor(denoised_image, cv2.COLOR_BGR2GRAY) # Convert to grayscale # Apply histogram equalization
|
163 |
-
detected_text = pytesseract.image_to_string(enhanced_image, config=custom_config)
|
164 |
-
# print(detected_text)
|
165 |
-
text+=detected_text
|
166 |
-
across_clues, down_clues = parse_crossword_clues(text)
|
167 |
-
return across_clues,down_clues
|
168 |
-
|
169 |
-
def get_text(image):
|
170 |
-
image = cv2.cvtColor(image,cv2.COLOR_GRAY2BGR)
|
171 |
-
result = first_preprocessing(image)
|
172 |
-
result1 = remove_head(result)
|
173 |
-
result2 = second_preprocessing(result1)
|
174 |
-
vertical_profile = find_vertical_profile(result2)
|
175 |
-
combined_columns = crop_text_columns(result2,vertical_profile)
|
176 |
-
across,down = classify_text(combined_columns)
|
177 |
-
return across,down
|
178 |
-
|
179 |
-
|
180 |
-
################################ Grid Extraction begins here ###########################
|
181 |
-
########################################################################################
|
182 |
-
|
183 |
-
|
184 |
-
# for applying non max suppression of the contours
|
185 |
-
def calculate_iou(image, contour1, contour2):
|
186 |
-
# Create masks for each contour
|
187 |
-
mask1 = np.zeros_like(image, dtype=np.uint8)
|
188 |
-
cv2.drawContours(mask1, [contour1], -1, 255, thickness=cv2.FILLED)
|
189 |
-
|
190 |
-
mask2 = np.zeros_like(image, dtype=np.uint8)
|
191 |
-
cv2.drawContours(mask2, [contour2], -1, 255, thickness=cv2.FILLED)
|
192 |
-
|
193 |
-
# Find the intersection between the two masks
|
194 |
-
intersection = cv2.bitwise_and(mask1, mask2)
|
195 |
-
|
196 |
-
# Calculate the intersection area
|
197 |
-
intersection_area = cv2.countNonZero(intersection)
|
198 |
-
|
199 |
-
# Calculate the union area (Not the accurate one but works alright XD !)
|
200 |
-
union_area = cv2.contourArea(cv2.convexHull(np.concatenate((contour1, contour2))))
|
201 |
-
|
202 |
-
# Calculate the IoU
|
203 |
-
iou = intersection_area / union_area
|
204 |
-
return iou
|
205 |
-
|
206 |
-
# remove overlapping contours, non square and not quardatic contours
|
207 |
-
# this check every contour with every other contour so be careful
|
208 |
-
def filter_contours(img_gray2, contours, iou_threshold = 0.6, asp_ratio = 1,tolerance = 0.5):
|
209 |
-
# Remove overlapping contours, removing that are not square
|
210 |
-
filtered_contours = []
|
211 |
-
epsilon = 0.02
|
212 |
-
for contour in contours:
|
213 |
-
|
214 |
-
# Approximate the contour to reduce the number of points
|
215 |
-
epsilon_multiplier = epsilon * cv2.arcLength(contour, True)
|
216 |
-
approximated_contour = cv2.approxPolyDP(contour, epsilon_multiplier, True)
|
217 |
-
|
218 |
-
# find the aspect ratio of the contour, if it is close to 1 then keep it otherwise discard
|
219 |
-
_,_,w,h = cv2.boundingRect(approximated_contour)
|
220 |
-
if(abs(float(w)/h - asp_ratio) > tolerance ): continue
|
221 |
-
|
222 |
-
# Calculate the IoU with all existing contours
|
223 |
-
iou_values = [calculate_iou(img_gray2,np.array(approximated_contour), np.array(existing_contour)) for existing_contour in filtered_contours]
|
224 |
-
|
225 |
-
# If the IoU value with all existing contours is below the threshold, add the current contour
|
226 |
-
if not any(iou_value > iou_threshold for iou_value in iou_values):
|
227 |
-
filtered_contours.append(approximated_contour)
|
228 |
-
|
229 |
-
return filtered_contours
|
230 |
-
|
231 |
-
# https://stackoverflow.com/questions/383480/intersection-of-two-lines-defined-in-rho-theta-parameterization/383527#383527
|
232 |
-
# Define the parametricIntersect function
|
233 |
-
def parametricIntersect(r1, t1, r2, t2):
|
234 |
-
ct1 = np.cos(t1)
|
235 |
-
st1 = np.sin(t1)
|
236 |
-
ct2 = np.cos(t2)
|
237 |
-
st2 = np.sin(t2)
|
238 |
-
d = ct1 * st2 - st1 * ct2
|
239 |
-
if d != 0.0:
|
240 |
-
x = int((st2 * r1 - st1 * r2) / d)
|
241 |
-
y = int((-ct2 * r1 + ct1 * r2) / d)
|
242 |
-
return x, y
|
243 |
-
else:
|
244 |
-
return None
|
245 |
-
|
246 |
-
# Group the coordinate to a list such that each point in a list may belong to a line
|
247 |
-
def group_lines(coordinates,axis=0,threshold=10):
|
248 |
-
sorted_coordinates = list(sorted(coordinates,key=lambda x: x[axis]))
|
249 |
-
groups = []
|
250 |
-
current_group = []
|
251 |
-
|
252 |
-
for i in range(len(sorted_coordinates)):
|
253 |
-
if i!=0 and abs(current_group[0][axis] - sorted_coordinates[i][axis]) > threshold: # condition to change the group
|
254 |
-
if len(current_group) > 4:
|
255 |
-
groups.append(current_group)
|
256 |
-
current_group = []
|
257 |
-
current_group.append(sorted_coordinates[i]) # condition to append to the group
|
258 |
-
if(len(current_group) > 4):
|
259 |
-
groups.append(current_group)
|
260 |
-
return groups
|
261 |
-
|
262 |
-
# Use the Grouped Lines to Fit a line using Linear Regression
|
263 |
-
def fit_lines(grouped_lines,is_horizontal = False):
|
264 |
-
actual_lines = []
|
265 |
-
for coordinates in grouped_lines:
|
266 |
-
# Converting into numpy array
|
267 |
-
coordinates_arr = np.array(coordinates)
|
268 |
-
# Separate the x and y coordinates
|
269 |
-
x = coordinates_arr[:, 0]
|
270 |
-
y = coordinates_arr[:, 1]
|
271 |
-
# Fit a linear regression model
|
272 |
-
regressor = LinearRegression()
|
273 |
-
regressor.fit(y.reshape(-1, 1), x)
|
274 |
-
# Get the slope and intercept of the fitted line
|
275 |
-
slope = regressor.coef_[0]
|
276 |
-
intercept = regressor.intercept_
|
277 |
-
|
278 |
-
if(is_horizontal):
|
279 |
-
intercept = np.mean(y)
|
280 |
-
actual_lines.append((slope,intercept))
|
281 |
-
|
282 |
-
return actual_lines
|
283 |
-
|
284 |
-
# Calculates difference between two consecutive elements in an array
|
285 |
-
def average_distance(arr):
|
286 |
-
n = len(arr)
|
287 |
-
distance_sum = 0
|
288 |
-
|
289 |
-
for i in range(n - 1):
|
290 |
-
distance_sum += abs(arr[i+1] - arr[i])
|
291 |
-
|
292 |
-
average = distance_sum / (n - 1)
|
293 |
-
return average
|
294 |
-
|
295 |
-
# If two adjacent lines are near than some threshold, then merge them
|
296 |
-
# Returns Results in y = mx + b from
|
297 |
-
def average_out_similar_lines(lines_m_c,lines_coord,del_threshold,is_horizontal=False):
|
298 |
-
averaged_lines = []
|
299 |
-
i = 0
|
300 |
-
while(i < len(lines_m_c) - 1):
|
301 |
-
|
302 |
-
_, intercept1 = lines_m_c[i]
|
303 |
-
_, intercept2 = lines_m_c[i + 1]
|
304 |
-
|
305 |
-
if abs(intercept2 - intercept1) < del_threshold:
|
306 |
-
new_points = np.array(lines_coord[i] + lines_coord[i+1][:-1])
|
307 |
-
# Separate the x and y coordinates
|
308 |
-
x = new_points[:, 0]
|
309 |
-
y = new_points[:, 1]
|
310 |
-
|
311 |
-
# Fit a linear regression model
|
312 |
-
regressor = LinearRegression()
|
313 |
-
regressor.fit(y.reshape(-1, 1), x)
|
314 |
-
|
315 |
-
# Get the slope and intercept of the fitted line
|
316 |
-
slope = regressor.coef_[0]
|
317 |
-
intercept = regressor.intercept_
|
318 |
-
|
319 |
-
if(is_horizontal):
|
320 |
-
intercept = np.mean(y)
|
321 |
-
averaged_lines.append((slope,intercept))
|
322 |
-
i+=2
|
323 |
-
else:
|
324 |
-
averaged_lines.append(lines_m_c[i])
|
325 |
-
i+=1
|
326 |
-
if(i < len(lines_m_c)):
|
327 |
-
averaged_lines.append(lines_m_c[i])
|
328 |
-
|
329 |
-
return averaged_lines
|
330 |
-
|
331 |
-
# If two adjacent lines are near than some threshold, then merge them
|
332 |
-
# Returns Results in normalized vector form
|
333 |
-
def average_out_similar_lines1(lines_m_c,lines_coord,del_threshold):
|
334 |
-
averaged_lines = []
|
335 |
-
i = 0
|
336 |
-
while(i < len(lines_m_c) - 1):
|
337 |
-
|
338 |
-
_, intercept1 = lines_m_c[i]
|
339 |
-
_, intercept2 = lines_m_c[i + 1]
|
340 |
-
|
341 |
-
if abs(intercept2 - intercept1) < del_threshold:
|
342 |
-
new_points = np.array(lines_coord[i] + lines_coord[i+1][:-1])
|
343 |
-
coordinates = np.array(new_points)
|
344 |
-
points = coordinates[:, None, :].astype(np.int32)
|
345 |
-
# Fit a line using linear regression
|
346 |
-
[vx, vy, x, y] = cv2.fitLine(points, cv2.DIST_L2, 0, 0.01, 0.01)
|
347 |
-
averaged_lines.append((vx, vy, x, y))
|
348 |
-
i+=2
|
349 |
-
else:
|
350 |
-
new_points = np.array(lines_coord[i])
|
351 |
-
|
352 |
-
coordinates = np.array(new_points)
|
353 |
-
points = coordinates[:, None, :].astype(np.int32)
|
354 |
-
# Fit a line using linear regression
|
355 |
-
[vx, vy, x, y] = cv2.fitLine(points, cv2.DIST_L2, 0, 0.01, 0.01)
|
356 |
-
averaged_lines.append((vx, vy, x, y))
|
357 |
-
i+=1
|
358 |
-
if(i < len(lines_m_c)):
|
359 |
-
new_points = np.array(lines_coord[i])
|
360 |
-
coordinates = np.array(new_points)
|
361 |
-
points = coordinates[:, None, :].astype(np.int32)
|
362 |
-
# Fit a line using linear regression
|
363 |
-
[vx, vy, x, y] = cv2.fitLine(points, cv2.DIST_L2, 0, 0.01, 0.01)
|
364 |
-
averaged_lines.append((vx, vy, x, y))
|
365 |
-
|
366 |
-
return averaged_lines
|
367 |
-
|
368 |
-
def get_square_color(image, box):
|
369 |
-
|
370 |
-
# Determine the size of the square region
|
371 |
-
square_size = (box[1][0] - box[0][0]) / 3
|
372 |
-
|
373 |
-
# Determine the coordinates of the square region inside the box
|
374 |
-
top_left = (box[0][0] + square_size, box[0][1] + square_size)
|
375 |
-
bottom_right = (box[0][0] + square_size*2, box[0][1] + square_size*2)
|
376 |
-
|
377 |
-
# Extract the square region from the image
|
378 |
-
square_region = image[int(top_left[1]):int(bottom_right[1]), int(top_left[0]):int(bottom_right[0])]
|
379 |
-
|
380 |
-
# Calculate the mean pixel value of the square region
|
381 |
-
mean_value = np.mean(square_region)
|
382 |
-
|
383 |
-
# Determine whether the square region is predominantly black or white
|
384 |
-
if mean_value < 128:
|
385 |
-
square_color = "."
|
386 |
-
else:
|
387 |
-
square_color = " "
|
388 |
-
|
389 |
-
return square_color
|
390 |
-
|
391 |
-
# accepts image in grayscale
|
392 |
-
def extract_grid(image):
|
393 |
-
|
394 |
-
# Apply Gaussian blur to reduce noise and improve edge detection
|
395 |
-
blurred = cv2.GaussianBlur(image, (3, 3), 0)
|
396 |
-
# Apply Canny edge detection
|
397 |
-
edges = cv2.Canny(blurred, 50, 150)
|
398 |
-
|
399 |
-
# Apply dilation to connect nearby edges and make them more contiguous
|
400 |
-
kernel = np.ones((5, 5), np.uint8)
|
401 |
-
dilated = cv2.dilate(edges, kernel, iterations=1)
|
402 |
-
|
403 |
-
# # Applying canny edge detector
|
404 |
-
# detecting contours on the canny image
|
405 |
-
contours, _ = cv2.findContours(dilated, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
|
406 |
-
|
407 |
-
# sorting the contours by the descending order area of the contour
|
408 |
-
sorted_contours = list(sorted(contours, key=cv2.contourArea,reverse=True))
|
409 |
-
# filtering out the top 10 largest by applying NMS and only selecting square ones (Apsect ratio 1)
|
410 |
-
filtered_contours = filter_contours(image, sorted_contours[0:10],iou_threshold=0.6,asp_ratio=1,tolerance=0.2)
|
411 |
-
|
412 |
-
# largest Contour Extraction
|
413 |
-
largest_contour = []
|
414 |
-
if(len(filtered_contours)):
|
415 |
-
largest_contour = filtered_contours[0]
|
416 |
-
else:
|
417 |
-
largest_contour = sorted_contours[0]
|
418 |
-
|
419 |
-
# --- Performing Perspective warp of the largest contour ---
|
420 |
-
coordinates_list = []
|
421 |
-
|
422 |
-
if(largest_contour.shape != (4,1,2)):
|
423 |
-
largest_contour = cv2.convexHull(largest_contour)
|
424 |
-
if(largest_contour.shape != (4,1,2)):
|
425 |
-
rect = cv2.minAreaRect(largest_contour)
|
426 |
-
largest_contour = cv2.boxPoints(rect)
|
427 |
-
largest_contour = largest_contour.astype('int')
|
428 |
-
|
429 |
-
coordinates_list = largest_contour.reshape(4, 2).tolist()
|
430 |
-
|
431 |
-
# Convert coordinates_list to a numpy array
|
432 |
-
coordinates_array = np.array(coordinates_list)
|
433 |
-
|
434 |
-
# Find the convex hull of the points
|
435 |
-
hull = cv2.convexHull(coordinates_array)
|
436 |
-
|
437 |
-
# Find the extreme points of the convex hull
|
438 |
-
extreme_points = np.squeeze(hull)
|
439 |
-
|
440 |
-
# Sort the extreme points by their x and y coordinates to determine the order
|
441 |
-
sorted_points = extreme_points[np.lexsort((extreme_points[:, 1], extreme_points[:, 0]))]
|
442 |
-
|
443 |
-
# Extract top left, bottom right, top right, and bottom left points
|
444 |
-
tl = sorted_points[0]
|
445 |
-
tr = sorted_points[1]
|
446 |
-
bl = sorted_points[2]
|
447 |
-
br = sorted_points[3]
|
448 |
-
|
449 |
-
if(tr[1] < tl[1]):
|
450 |
-
tl,tr = tr,tl
|
451 |
-
if(br[1] < bl[1]):
|
452 |
-
bl,br = br,bl
|
453 |
-
|
454 |
-
# Define pts1
|
455 |
-
pts1 = [tl, bl, tr, br]
|
456 |
-
|
457 |
-
# Calculate the bounding rectangle coordinates
|
458 |
-
x, y, w, h = 0,0,400,400
|
459 |
-
# Define pts2 as the corners of the bounding rectangle
|
460 |
-
pts2 = [[3, 3], [400, 3], [3, 400], [400, 400]]
|
461 |
-
|
462 |
-
# Calculate the perspective transformation matrix
|
463 |
-
matrix = cv2.getPerspectiveTransform(np.float32(pts1), np.float32(pts2))
|
464 |
-
|
465 |
-
# Apply the perspective transformation to the cropped_image
|
466 |
-
transformed_img = cv2.warpPerspective(image, matrix, (403, 403))
|
467 |
-
cropped_image = transformed_img.copy()
|
468 |
-
|
469 |
-
# if the largest contour was not exactly quadilateral
|
470 |
-
|
471 |
-
# -- Performing Hough Transform --
|
472 |
-
|
473 |
-
similarity_threshold = math.floor(w/30) # Thresholds for filtering Similar Hough Lines
|
474 |
-
|
475 |
-
# Applying Gaussian Blur to reduce noice and improve dege detection
|
476 |
-
blurred = cv2.GaussianBlur(cropped_image, (5, 5), 0)
|
477 |
-
# Perform Canny edge detection on the GrayScale Image
|
478 |
-
edges = cv2.Canny(blurred, 50, 150)
|
479 |
-
lines = cv2.HoughLines(edges, 1, np.pi/180, 200)
|
480 |
-
|
481 |
-
# Filter out similar lines
|
482 |
-
filtered_lines = []
|
483 |
-
for line in lines:
|
484 |
-
for r_theta in lines:
|
485 |
-
arr = np.array(r_theta[0], dtype=np.float64)
|
486 |
-
rho, theta = arr
|
487 |
-
is_similar = False
|
488 |
-
for filtered_line in filtered_lines:
|
489 |
-
filtered_rho, filtered_theta = filtered_line
|
490 |
-
# similarity threshold is 10
|
491 |
-
if abs(rho - filtered_rho) < similarity_threshold and abs(theta - filtered_theta) < np.pi/180 * similarity_threshold:
|
492 |
-
is_similar = True
|
493 |
-
break
|
494 |
-
if not is_similar:
|
495 |
-
filtered_lines.append((rho, theta))
|
496 |
-
|
497 |
-
# Filter out the horizontal and the vertical lines
|
498 |
-
horizontal_lines = []
|
499 |
-
vertical_lines = []
|
500 |
-
for rho, theta in filtered_lines:
|
501 |
-
a = np.cos(theta)
|
502 |
-
b = np.sin(theta)
|
503 |
-
x0 = a * rho
|
504 |
-
y0 = b * rho
|
505 |
-
x1 = int(x0 + 1000 * (-b))
|
506 |
-
y1 = int(y0 + 1000 * (a))
|
507 |
-
x2 = int(x0 - 1000 * (-b))
|
508 |
-
y2 = int(y0 - 1000 * (a))
|
509 |
-
|
510 |
-
slope = (y2 - y1) / (x2 - x1 + 0.0001)
|
511 |
-
# do taninv(0.17) it is nearly equal to 10
|
512 |
-
if( abs(slope) <= 0.18 ):
|
513 |
-
horizontal_lines.append((rho,theta))
|
514 |
-
elif (abs(slope) > 6):
|
515 |
-
vertical_lines.append((rho,theta))
|
516 |
-
|
517 |
-
# Find the intersection points of horizontal and vertical lines
|
518 |
-
hough_corners = []
|
519 |
-
for h_rho, h_theta in horizontal_lines:
|
520 |
-
for v_rho, v_theta in vertical_lines:
|
521 |
-
x, y = parametricIntersect(h_rho, h_theta, v_rho, v_theta)
|
522 |
-
if x is not None and y is not None:
|
523 |
-
hough_corners.append((x, y))
|
524 |
-
|
525 |
-
# -- Performing Harris Corner Detection --
|
526 |
-
|
527 |
-
# Create CLAHE object with specified clip limit
|
528 |
-
clahe = cv2.createCLAHE(clipLimit=3, tileGridSize=(8, 8))
|
529 |
-
clahe_image = clahe.apply(cropped_image)
|
530 |
-
|
531 |
-
# harris corner detection for CLHAE IMAGE
|
532 |
-
dst = cv2.cornerHarris(clahe_image,2,3,0.04)
|
533 |
-
ret,dst = cv2.threshold(dst,0.1*dst.max(),255,0)
|
534 |
-
dst = np.uint8(dst)
|
535 |
-
dst = cv2.dilate(dst,None)
|
536 |
-
ret, labels, stats, centroids = cv2.connectedComponentsWithStats(dst)
|
537 |
-
criteria = (cv2.TERM_CRITERIA_EPS+cv2.TermCriteria_MAX_ITER,100,0.001)
|
538 |
-
harris_corners = cv2.cornerSubPix(clahe_image,np.float32(centroids),(5,5),(-1,-1),criteria)
|
539 |
-
|
540 |
-
drawn_image = cv2.cvtColor(cropped_image, cv2.COLOR_GRAY2BGR)
|
541 |
-
for i in harris_corners:
|
542 |
-
x,y = i
|
543 |
-
image2 = cv2.circle(drawn_image, (int(x),int(y)), radius=0, color=(0, 0, 255), thickness=3)
|
544 |
-
|
545 |
-
# -- Using Regression Model to approximate horizontal and vertical Lines
|
546 |
-
|
547 |
-
# reducing to 0 decimal places
|
548 |
-
corners1 = list(map(lambda coord: (round(coord[0], 0), round(coord[1], 0)), harris_corners))
|
549 |
-
|
550 |
-
# adding the corners obtained from hough transform
|
551 |
-
corners1 += hough_corners
|
552 |
-
|
553 |
-
# removing the duplicate corners
|
554 |
-
corners_no_dup = list(set(corners1))
|
555 |
-
|
556 |
-
min_cell_width = w/30
|
557 |
-
min_cell_height = h/30
|
558 |
-
|
559 |
-
# grouping coordinates into probabale array that could fit a horizontal and vertical lien
|
560 |
-
vertical_lines = group_lines(corners_no_dup,0,min_cell_height)
|
561 |
-
horizontal_lines = group_lines(corners_no_dup,1,min_cell_height)
|
562 |
-
|
563 |
-
actual_vertical_lines = fit_lines(vertical_lines)
|
564 |
-
actual_horizontal_lines = fit_lines(horizontal_lines,is_horizontal=True)
|
565 |
-
|
566 |
-
|
567 |
-
# Lines obtained from above method are not appropriate, we have to refine them
|
568 |
-
|
569 |
-
x_probable = [i[1] for i in actual_horizontal_lines] # looking at the intercepts
|
570 |
-
y_probable = [i[1] for i in actual_vertical_lines]
|
571 |
-
|
572 |
-
del_x_avg = average_distance(x_probable)
|
573 |
-
del_y_avg = average_distance(y_probable)
|
574 |
-
|
575 |
-
averaged_horizontal_lines1 = [] # This step here is fishy and needs refinement
|
576 |
-
averaged_vertical_lines1 = []
|
577 |
-
multiplier = 0.95
|
578 |
-
i = 0
|
579 |
-
while(1):
|
580 |
-
averaged_horizontal_lines = average_out_similar_lines(actual_horizontal_lines,horizontal_lines,del_y_avg*multiplier,is_horizontal=True)
|
581 |
-
averaged_vertical_lines = average_out_similar_lines(actual_vertical_lines,vertical_lines,del_x_avg*multiplier,is_horizontal=False)
|
582 |
-
i += 1
|
583 |
-
if(i >= 20 or len(averaged_horizontal_lines) == len(averaged_vertical_lines)):
|
584 |
-
break
|
585 |
-
else:
|
586 |
-
multiplier -= 0.05
|
587 |
-
|
588 |
-
averaged_horizontal_lines1 = average_out_similar_lines1(actual_horizontal_lines,horizontal_lines,del_y_avg*multiplier)
|
589 |
-
averaged_vertical_lines1 = average_out_similar_lines1(actual_vertical_lines,vertical_lines,del_x_avg*multiplier)
|
590 |
-
|
591 |
-
|
592 |
-
# plotting the lines to image to find the intersection points
|
593 |
-
drawn_image6 = np.ones_like(cropped_image)*255
|
594 |
-
for vx,vy,cx,cy in averaged_horizontal_lines1 + averaged_vertical_lines1:
|
595 |
-
w = cropped_image.shape[1]
|
596 |
-
cv2.line(drawn_image6, (int(cx-vx*w), int(cy-vy*w)), (int(cx+vx*w), int(cy+vy*w)), (0, 0, 255),1,cv2.LINE_AA)
|
597 |
-
|
598 |
-
# -- Finding Intersection points --
|
599 |
-
|
600 |
-
# Applying Harris Corner Detection to find the intersection points
|
601 |
-
mesh_image = drawn_image6.copy()
|
602 |
-
dst = cv2.cornerHarris(mesh_image,2,3,0.04)
|
603 |
-
|
604 |
-
ret,dst = cv2.threshold(dst,0.1*dst.max(),255,0)
|
605 |
-
dst = np.uint8(dst)
|
606 |
-
dst = cv2.dilate(dst,None)
|
607 |
-
ret, labels, stats, centroids = cv2.connectedComponentsWithStats(dst)
|
608 |
-
criteria = (cv2.TERM_CRITERIA_EPS+cv2.TermCriteria_MAX_ITER,100,0.001)
|
609 |
-
harris_corners = cv2.cornerSubPix(mesh_image,np.float32(centroids),(5,5),(-1,-1),criteria)
|
610 |
-
drawn_image = cv2.cvtColor(drawn_image6, cv2.COLOR_GRAY2BGR)
|
611 |
-
harris_corners = list(sorted(harris_corners[1:],key = lambda x : x[1]))
|
612 |
-
|
613 |
-
# -- Finding out the grid color --
|
614 |
-
|
615 |
-
|
616 |
-
grayscale = cropped_image.copy()
|
617 |
-
# Perform adaptive thresholding to obtain binary image
|
618 |
-
_, binary = cv2.threshold(grayscale, 128, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)
|
619 |
-
|
620 |
-
# Perform morphological operations to remove small text regions
|
621 |
-
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
|
622 |
-
binary = cv2.morphologyEx(binary, cv2.MORPH_ELLIPSE, kernel, iterations=1)
|
623 |
-
|
624 |
-
# Invert the binary image
|
625 |
-
inverted_binary = cv2.bitwise_not(binary)
|
626 |
-
|
627 |
-
# Restore the image by blending the inverted binary image with the grayscale image
|
628 |
-
restored_image = cv2.bitwise_or(inverted_binary, grayscale)
|
629 |
-
|
630 |
-
# Apply morphological opening to remove small black dots
|
631 |
-
kernel_opening = np.ones((3, 3), np.uint8)
|
632 |
-
opened_image = cv2.morphologyEx(restored_image, cv2.MORPH_OPEN, kernel_opening, iterations=1)
|
633 |
-
|
634 |
-
# Apply morphological closing to further refine the restored image
|
635 |
-
kernel_closing = np.ones((5, 5), np.uint8)
|
636 |
-
refined_image = cv2.morphologyEx(opened_image, cv2.MORPH_CLOSE, kernel_closing, iterations=1)
|
637 |
-
|
638 |
-
# finding out the grid corner
|
639 |
-
grid = []
|
640 |
-
grid_nums = []
|
641 |
-
across_clue_num = []
|
642 |
-
down_clue_num = []
|
643 |
-
|
644 |
-
sorted_corners = np.array(list(sorted(harris_corners,key=lambda x:x[1])))
|
645 |
-
if(len(sorted_corners) == len(averaged_horizontal_lines1) * len(averaged_vertical_lines1)):
|
646 |
-
sorted_corners_grouped = []
|
647 |
-
for i in range(0,len(sorted_corners),len(averaged_vertical_lines1)):
|
648 |
-
temp_arr = sorted_corners[i:i+len(averaged_vertical_lines1)]
|
649 |
-
temp_arr = list(sorted(temp_arr,key=lambda x: x[0]))
|
650 |
-
sorted_corners_grouped.append(temp_arr)
|
651 |
-
|
652 |
-
for h_line_idx in range(0,len(sorted_corners_grouped)-1):
|
653 |
-
for corner_idx in range(0,len(sorted_corners_grouped[h_line_idx])-1):
|
654 |
-
# grabbing the four box coordinates
|
655 |
-
box = [sorted_corners_grouped[h_line_idx][corner_idx],sorted_corners_grouped[h_line_idx][corner_idx+1],
|
656 |
-
sorted_corners_grouped[h_line_idx+1][corner_idx],sorted_corners_grouped[h_line_idx+1][corner_idx+1]]
|
657 |
-
grid.append(get_square_color(refined_image,box))
|
658 |
-
|
659 |
-
grid_formatted = []
|
660 |
-
for i in range(0, len(grid), len(averaged_vertical_lines1) - 1):
|
661 |
-
grid_formatted.append(grid[i:i + len(averaged_vertical_lines1) - 1])
|
662 |
-
|
663 |
-
|
664 |
-
# if (x,y) is present in these array the cell (x,y) is already accounted as a part of answer of across or down
|
665 |
-
in_horizontal = []
|
666 |
-
in_vertical = []
|
667 |
-
|
668 |
-
num = 0
|
669 |
-
|
670 |
-
|
671 |
-
|
672 |
-
for x in range(0, len(averaged_vertical_lines1) - 1):
|
673 |
-
for y in range(0, len(averaged_horizontal_lines1) - 1):
|
674 |
-
|
675 |
-
# if the cell is black there's no need to number
|
676 |
-
if grid_formatted[x][y] == '.':
|
677 |
-
grid_nums.append(0)
|
678 |
-
continue
|
679 |
-
|
680 |
-
# if the cell is part of both horizontal and vertical cell then there's no need to number
|
681 |
-
horizontal_presence = (x, y) in in_horizontal
|
682 |
-
vertical_presence = (x, y) in in_vertical
|
683 |
-
|
684 |
-
# present in both 1 1
|
685 |
-
if horizontal_presence and vertical_presence:
|
686 |
-
grid_nums.append(0)
|
687 |
-
continue
|
688 |
-
|
689 |
-
# present in one i.e 1 0
|
690 |
-
if not horizontal_presence and vertical_presence:
|
691 |
-
horizontal_length = 0
|
692 |
-
temp_horizontal_arr = []
|
693 |
-
# iterate in x direction until the end of the grid or until a black box is found
|
694 |
-
while x + horizontal_length < len(averaged_horizontal_lines1) - 1 and grid_formatted[x + horizontal_length][y] != '.':
|
695 |
-
temp_horizontal_arr.append((x + horizontal_length, y))
|
696 |
-
horizontal_length += 1
|
697 |
-
# if horizontal length is greater than 1, then append the temp_horizontal_arr to in_horizontal array
|
698 |
-
if horizontal_length > 1:
|
699 |
-
in_horizontal.extend(temp_horizontal_arr)
|
700 |
-
num += 1
|
701 |
-
across_clue_num.append(num)
|
702 |
-
grid_nums.append(num)
|
703 |
-
continue
|
704 |
-
grid_nums.append(0)
|
705 |
-
# present in one 1 0
|
706 |
-
if not vertical_presence and horizontal_presence:
|
707 |
-
# do the same for vertical
|
708 |
-
vertical_length = 0
|
709 |
-
temp_vertical_arr = []
|
710 |
-
# iterate in y direction until the end of the grid or until a black box is found
|
711 |
-
while y + vertical_length < len(averaged_vertical_lines1) - 1 and grid_formatted[x][y+vertical_length] != '.':
|
712 |
-
temp_vertical_arr.append((x, y+vertical_length))
|
713 |
-
vertical_length += 1
|
714 |
-
# if vertical length is greater than 1, then append the temp_vertical_arr to in_vertical array
|
715 |
-
if vertical_length > 1:
|
716 |
-
in_vertical.extend(temp_vertical_arr)
|
717 |
-
num += 1
|
718 |
-
down_clue_num.append(num)
|
719 |
-
grid_nums.append(num)
|
720 |
-
continue
|
721 |
-
grid_nums.append(0)
|
722 |
-
|
723 |
-
if(not horizontal_presence and not vertical_presence):
|
724 |
-
|
725 |
-
horizontal_length = 0
|
726 |
-
temp_horizontal_arr = []
|
727 |
-
# iterate in x direction until the end of the grid or until a black box is found
|
728 |
-
while x + horizontal_length < len(averaged_horizontal_lines1) - 1 and grid_formatted[x + horizontal_length][y] != '.':
|
729 |
-
temp_horizontal_arr.append((x + horizontal_length, y))
|
730 |
-
horizontal_length += 1
|
731 |
-
# if horizontal length is greater than 1, then append the temp_horizontal_arr to in_horizontal array
|
732 |
-
|
733 |
-
# do the same for vertical
|
734 |
-
vertical_length = 0
|
735 |
-
temp_vertical_arr = []
|
736 |
-
# iterate in y direction until the end of the grid or until a black box is found
|
737 |
-
while y + vertical_length < len(averaged_vertical_lines1) - 1 and grid_formatted[x][y+vertical_length] != '.':
|
738 |
-
temp_vertical_arr.append((x, y+vertical_length))
|
739 |
-
vertical_length += 1
|
740 |
-
# if vertical length is greater than 1, then append the temp_vertical_arr to in_vertical array
|
741 |
-
|
742 |
-
if horizontal_length > 1 and horizontal_length > 1:
|
743 |
-
in_horizontal.extend(temp_horizontal_arr)
|
744 |
-
in_vertical.extend(temp_vertical_arr)
|
745 |
-
num += 1
|
746 |
-
across_clue_num.append(num)
|
747 |
-
down_clue_num.append(num)
|
748 |
-
grid_nums.append(num)
|
749 |
-
elif vertical_length > 1:
|
750 |
-
in_vertical.extend(temp_vertical_arr)
|
751 |
-
num += 1
|
752 |
-
down_clue_num.append(num)
|
753 |
-
grid_nums.append(num)
|
754 |
-
elif horizontal_length > 1:
|
755 |
-
in_horizontal.extend(temp_horizontal_arr)
|
756 |
-
num += 1
|
757 |
-
across_clue_num.append(num)
|
758 |
-
grid_nums.append(num)
|
759 |
-
else:
|
760 |
-
grid_nums.append(0)
|
761 |
-
|
762 |
-
|
763 |
-
size = { 'rows' : len(averaged_horizontal_lines1)-1,
|
764 |
-
'cols' : len(averaged_vertical_lines1)-1,
|
765 |
-
}
|
766 |
-
|
767 |
-
dict = {
|
768 |
-
'size' : size,
|
769 |
-
'grid' : grid,
|
770 |
-
'gridnums': grid_nums,
|
771 |
-
'across_nums': down_clue_num,
|
772 |
-
'down_nums' : across_clue_num,
|
773 |
-
'clues':{
|
774 |
-
'across' : [],
|
775 |
-
'down': []
|
776 |
-
}
|
777 |
-
}
|
778 |
-
|
779 |
-
return dict
|
780 |
-
|
781 |
-
if __name__ == "__main__":
|
782 |
-
img = cv2.imread("D:\\D\\Major Project files\\opencv\\movie.png",0)
|
783 |
-
down = extract_grid(img)
|
784 |
-
print(down)
|
785 |
-
# img = Image.open("chalena3.jpg")
|
786 |
-
# img_gray = img.convert("L")
|
787 |
-
# print(extract_grid(img_gray))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
main.py
CHANGED
@@ -1,26 +1,17 @@
|
|
1 |
-
from fastapi import FastAPI,
|
2 |
import os
|
3 |
from Crossword_inf import Crossword
|
4 |
from BPSolver_inf import BPSolver
|
5 |
from Strict_json import json_CA_json_converter
|
6 |
|
|
|
7 |
from fastapi.middleware.cors import CORSMiddleware
|
8 |
-
import aiofiles
|
9 |
-
import cv2
|
10 |
-
from extractpuzzle import extract_grid,get_text
|
11 |
-
|
12 |
-
MODEL_PATH = os.path.join("Inference_components","dpr_biencoder_trained_500k.bin")
|
13 |
-
ANS_TSV_PATH = os.path.join("Inference_components","all_answer_list.tsv")
|
14 |
-
DENSE_EMBD_PATH = os.path.join("Inference_components","embeddings_all_answers_json_0*")
|
15 |
|
16 |
MODEL_PATH_DISTIL = os.path.join("Inference_components","distilbert_EPOCHs_7_COMPLETE.bin")
|
17 |
ANS_TSV_PATH_DISTIL = os.path.join("Inference_components","all_answer_list.tsv")
|
18 |
DENSE_EMBD_PATH_DISTIL = os.path.join("Inference_components","distilbert_7_epochs_embeddings.pkl")
|
19 |
|
20 |
-
|
21 |
app = FastAPI()
|
22 |
-
# for reading images in chunk
|
23 |
-
CHUNK_SIZE = 1024 * 1024 * 2
|
24 |
|
25 |
app.add_middleware(
|
26 |
CORSMiddleware,
|
@@ -29,65 +20,75 @@ app.add_middleware(
|
|
29 |
allow_headers=["*"],
|
30 |
allow_credentials=True,
|
31 |
)
|
32 |
-
|
33 |
-
@app.get("/")
|
34 |
-
async def index():
|
35 |
-
return {"message": "Hello World"}
|
36 |
-
|
37 |
-
@app.post("/solve")
|
38 |
-
async def solve(request: Request):
|
39 |
-
json = await request.json()
|
40 |
-
puzzle = json_CA_json_converter(json, False)
|
41 |
-
crossword = Crossword(puzzle)
|
42 |
-
solver = BPSolver(crossword, model_path = MODEL_PATH_DISTIL,
|
43 |
-
ans_tsv_path = ANS_TSV_PATH_DISTIL,
|
44 |
-
dense_embd_path = DENSE_EMBD_PATH_DISTIL,
|
45 |
-
max_candidates = 40000,
|
46 |
-
model_type = 'distilbert')
|
47 |
-
solution = solver.solve(num_iters = 100, iterative_improvement_steps = 0)
|
48 |
-
return solution, solver.evaluate(solution)
|
49 |
-
|
50 |
-
@app.post("/parseImage/")
|
51 |
-
async def upload(file: UploadFile = File(...)):
|
52 |
-
|
53 |
-
try:
|
54 |
-
filepath = os.path.join('./', os.path.basename(file.filename))
|
55 |
-
async with aiofiles.open(filepath, 'wb') as f:
|
56 |
-
while chunk := await file.read(CHUNK_SIZE):
|
57 |
-
await f.write(chunk)
|
58 |
-
except Exception:
|
59 |
-
raise HTTPException(status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
|
60 |
-
detail='There was an error uploading the file')
|
61 |
-
finally:
|
62 |
-
await file.close()
|
63 |
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
|
|
|
|
|
|
|
|
|
69 |
|
70 |
-
try: # try extracting the grid from the image
|
71 |
-
# dict = { 'size' : size, 'grid' : grid, 'gridnums': grid_nums, 'across_nums': down_clue_num,'down_nums' : across_clue_num }
|
72 |
-
grid_data = extract_grid(img_array)
|
73 |
-
grid_data['gridExtractionStatus'] = "Passed"
|
74 |
-
except Exception as e:
|
75 |
-
grid_data['gridExtractionStatus'] = "Failed"
|
76 |
|
77 |
-
|
78 |
-
try: # try extracting clues
|
79 |
-
acrossClues, downClues = get_text(img_array) # { number : [column_of_projection_profile,extracted_text]}
|
80 |
-
clue_data['across'] = acrossClues
|
81 |
-
clue_data['down'] = downClues
|
82 |
-
clue_data['gridExtractionStatus'] = "Passed"
|
83 |
-
except Exception as e:
|
84 |
-
grid_data['ClueExtractionStatus'] = "Failed"
|
85 |
|
86 |
-
|
87 |
|
88 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
|
90 |
-
|
|
|
|
|
91 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
92 |
|
|
|
|
|
|
|
|
|
|
|
|
|
93 |
|
|
|
1 |
+
from fastapi import FastAPI,Request
|
2 |
import os
|
3 |
from Crossword_inf import Crossword
|
4 |
from BPSolver_inf import BPSolver
|
5 |
from Strict_json import json_CA_json_converter
|
6 |
|
7 |
+
import asyncio
|
8 |
from fastapi.middleware.cors import CORSMiddleware
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
MODEL_PATH_DISTIL = os.path.join("Inference_components","distilbert_EPOCHs_7_COMPLETE.bin")
|
11 |
ANS_TSV_PATH_DISTIL = os.path.join("Inference_components","all_answer_list.tsv")
|
12 |
DENSE_EMBD_PATH_DISTIL = os.path.join("Inference_components","distilbert_7_epochs_embeddings.pkl")
|
13 |
|
|
|
14 |
app = FastAPI()
|
|
|
|
|
15 |
|
16 |
app.add_middleware(
|
17 |
CORSMiddleware,
|
|
|
20 |
allow_headers=["*"],
|
21 |
allow_credentials=True,
|
22 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
|
24 |
+
async def solve_puzzle(json):
|
25 |
+
puzzle = json_CA_json_converter(json, False)
|
26 |
+
crossword = Crossword(puzzle)
|
27 |
+
|
28 |
+
# Perform asynchronous operations using asyncio.gather or asyncio.create_task
|
29 |
+
async def solve_async():
|
30 |
+
return await asyncio.to_thread(BPSolver, crossword, model_path=MODEL_PATH_DISTIL,
|
31 |
+
ans_tsv_path=ANS_TSV_PATH_DISTIL,
|
32 |
+
dense_embd_path=DENSE_EMBD_PATH_DISTIL,
|
33 |
+
max_candidates=40000,
|
34 |
+
model_type='distilbert')
|
35 |
+
|
36 |
+
solver = await solve_async()
|
37 |
+
|
38 |
+
# Run solve method asynchronously
|
39 |
+
async def solve_method_async():
|
40 |
+
return await asyncio.to_thread(solver.solve, num_iters=100, iterative_improvement_steps=0)
|
41 |
|
42 |
+
solution = await solve_method_async()
|
43 |
+
evaluation = await asyncio.to_thread(solver.evaluate, solution)
|
44 |
+
|
45 |
+
return solution, evaluation
|
46 |
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
|
48 |
+
fifo_queue = asyncio.Queue()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
49 |
|
50 |
+
jobs = {}
|
51 |
|
52 |
+
async def worker():
|
53 |
+
while True:
|
54 |
+
print(f"Worker got a job: (size of remaining queue: {fifo_queue.qsize()})")
|
55 |
+
job_id, job, args, future = await fifo_queue.get()
|
56 |
+
jobs[job_id]["status"] = "processing"
|
57 |
+
result = await job(*args)
|
58 |
+
jobs[job_id]["result"] = result
|
59 |
+
jobs[job_id]["status"] = "completed"
|
60 |
+
future.set_result(job_id)
|
61 |
|
62 |
+
@app.on_event("startup")
|
63 |
+
async def on_start_up():
|
64 |
+
asyncio.create_task(worker())
|
65 |
|
66 |
+
@app.post("/solve")
|
67 |
+
async def solve(request: Request):
|
68 |
+
json = await request.json()
|
69 |
+
future = asyncio.Future()
|
70 |
+
job_id = id(future)
|
71 |
+
jobs[job_id]= {"status":"queued"}
|
72 |
+
await fifo_queue.put((job_id, solve_puzzle, [json], future))
|
73 |
+
return {"job_id": job_id}
|
74 |
+
|
75 |
+
@app.get("/result/{job_id}")
|
76 |
+
async def get_result(job_id: int):
|
77 |
+
if job_id in jobs:
|
78 |
+
returnVal = {}
|
79 |
+
returnVal = {**jobs[job_id]}
|
80 |
+
if(jobs[job_id]["status"]=="queued"):
|
81 |
+
queue_size = fifo_queue.qsize()
|
82 |
+
for index, (queued_job_id, _, _, _) in enumerate(fifo_queue._queue):
|
83 |
+
if job_id == queued_job_id:
|
84 |
+
returnVal["queue_status"] = f"Enqueued In : {index + 1}/{queue_size}"
|
85 |
+
return returnVal
|
86 |
+
return {"error": "Job not found or completed"}
|
87 |
|
88 |
+
@app.get("/")
|
89 |
+
async def home():
|
90 |
+
return {
|
91 |
+
"Success" : "True",
|
92 |
+
"Message" : "Pong"
|
93 |
+
}
|
94 |
|
requirements.txt
CHANGED
@@ -7,9 +7,7 @@ transformers==4.35.2
|
|
7 |
wordsegment==1.3.1
|
8 |
torch==2.1.1
|
9 |
faiss-cpu==1.7.4
|
10 |
-
aiofiles==23.2.1
|
11 |
python-multipart
|
12 |
-
opencv-python-headless==4.6.0.66
|
13 |
pytesseract==0.3.10
|
14 |
scikit-learn==1.3.2
|
15 |
|
|
|
7 |
wordsegment==1.3.1
|
8 |
torch==2.1.1
|
9 |
faiss-cpu==1.7.4
|
|
|
10 |
python-multipart
|
|
|
11 |
pytesseract==0.3.10
|
12 |
scikit-learn==1.3.2
|
13 |
|