-
Notifications
You must be signed in to change notification settings - Fork 2
/
segment.py
291 lines (257 loc) · 9.85 KB
/
segment.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
from PIL import Image
import numpy as np
import csv
import matplotlib.pyplot as plt
# from skimage import color
import os
def rgb_dist(rgb1, rgb2):
return (int(rgb1[0])-int(rgb2[0]))**2+(int(rgb1[1])-int(rgb2[1]))**2+(int(rgb1[2])-int(rgb2[2]))**2
def hex_dist(hex1, hex2):
return rgb_dist(hex2rgb(hex1),hex2rgb(hex2))
def get_nearest_hex(rgb_hex,palette):
#returns nearest color in hex
dist = [hex_dist(rgb_hex, p_hex) for p_hex in palette]
return palette[dist.index(min(dist))]
def get_nearest_rgb(rgb, palette):
return get_nearest_hex(rgb2hex((rgb[0],rgb[1],rgb[2])), palette)
def rgb2hex(rgb):
return ('%02x%02x%02x' % rgb).upper()
def hex2rgb(value):
value = value.lstrip('#')
lv = len(value)
return tuple(int(value[i:i+lv//3], 16) for i in range(0, lv, lv//3))
def get_color(img,i,j,palette):
#gets hex value in palette of nearest color in 8 neighboring pixels
h,w,d = img.shape
c1 = img[i][j]
c1hex = rgb2hex((c1[0],c1[1],c1[2]))
if c1hex in palette:
return c1hex
colors = []
distances = []
for a in range(-1,2):
for b in range(-1,2):
if (a == 0 and b == 0) or i+a < 0 or i+a >= h or j+b < 0 or j+b >= w :
continue
c2 = img[i+a][j+b]
# if rgb2hex((c2[0],c2[1],c2[2])) not in palette:
# continue
colors.append(c2)
d = rgb_dist(c1,c2)
distances.append(d)
# if len(distances) == 0:
# return get_nearest_hex(c1hex,palette)
return get_nearest_rgb(colors[distances.index(min(distances))],palette)
#return rgb2hex((n_rgb[0],n_rgb[1],n_rgb[2]))
# return get_nearest_hex(rgb2hex((n_rgb[0],n_rgb[1],n_rgb[2])),palette)
def getsegment(img,i,j,palette,visited,px2id,adjacency,seg_id):
#do bfs to get image segment
h,w,d = img.shape
# col = get_nearest_rgb(img[i][j],palette)
col = get_color(img,i,j,palette)
segment = set()
q = []
q.append((i,j))
visited.add((i,j))
while len(q) > 0:
pi,pj = q.pop(0)
segment.add((pi,pj))
px2id[pi][pj] = seg_id
# if pi > 0 and (pi-1,pj) not in visited and get_nearest_rgb(img[pi-1][pj],palette) == col:
if pi > 0:
if (pi-1,pj) not in visited and get_color(img,pi-1,pj,palette) == col:
q.append((pi-1,pj))
visited.add((pi-1,pj))
elif px2id[pi-1][pj] != -1 and px2id[pi-1][pj] != seg_id:
adj_id = px2id[pi-1][pj]
if seg_id not in adjacency:
adjacency[seg_id] = set()
if adj_id not in adjacency:
adjacency[adj_id] = set()
adjacency[seg_id].add(adj_id)
adjacency[adj_id].add(seg_id)
# if pi+1 < h and (pi+1,pj) not in visited and get_nearest_rgb(img[pi+1][pj],palette) == col:
if pi+1 < h:
if (pi+1,pj) not in visited and get_color(img,pi+1,pj,palette) == col:
q.append((pi+1,pj))
visited.add((pi+1,pj))
elif px2id[pi+1][pj] != -1 and px2id[pi+1][pj] != seg_id:
adj_id = px2id[pi+1][pj]
if seg_id not in adjacency:
adjacency[seg_id] = set()
if adj_id not in adjacency:
adjacency[adj_id] = set()
adjacency[seg_id].add(adj_id)
adjacency[adj_id].add(seg_id)
# if pj > 0 and (pi,pj-1) not in visited and get_nearest_rgb(img[pi][pj-1],palette) == col:
if pj > 0:
if (pi,pj-1) not in visited and get_color(img,pi,pj-1,palette) == col:
q.append((pi,pj-1))
visited.add((pi,pj-1))
elif px2id[pi][pj-1] != -1 and px2id[pi][pj-1] != seg_id:
adj_id = px2id[pi][pj-1]
if seg_id not in adjacency:
adjacency[seg_id] = set()
if adj_id not in adjacency:
adjacency[adj_id] = set()
adjacency[seg_id].add(adj_id)
adjacency[adj_id].add(seg_id)
# if pj+1 < w and (pi,pj+1) not in visited and get_nearest_rgb(img[pi][pj+1],palette) == col:
if pj+1 < w:
if (pi,pj+1) not in visited and get_color(img,pi,pj+1,palette) == col:
q.append((pi,pj+1))
visited.add((pi,pj+1))
elif px2id[pi][pj+1] != -1 and px2id[pi][pj+1] != seg_id:
adj_id = px2id[pi][pj+1]
if seg_id not in adjacency:
adjacency[seg_id] = set()
if adj_id not in adjacency:
adjacency[adj_id] = set()
adjacency[seg_id].add(adj_id)
adjacency[adj_id].add(seg_id)
return list(segment)
# matrix = pixel to segment_id
def enclosure_strengths(matrix, num_ids, adjacency):
n = len(matrix)
m = len(matrix[0])
dist = 2
count = [[0 for i in range(num_ids+1)] for j in range(num_ids)] # row segment, col is neighboring segment, value is unnnormalized strength
for i in range(-dist, n+dist):
for j in range(-dist, m+dist):
s = set()
outofbounds = set()
for dx in range(-dist, dist+1):
for dy in range(-dist, dist+1):
nx, ny = i+dx, j+dy
if nx >= 0 and nx < n and ny >= 0 and ny < m:
if i >= 0 and i < n and j >= 0 and j < m:
if matrix[i][j] != matrix[nx][ny]:
s.add(matrix[nx][ny])
else:
outofbounds.add(matrix[nx][ny])
for k in s:
count[k][matrix[i][j]]+=1
for k in outofbounds:
count[k][num_ids] += 1
for i in range(len(count)):
for j in range(len(count[0])):
if count[i][j] != 0:
if not(j in adjacency[i]):
count[i][j] = 0
# normalize
for i in range(len(count)):
total = sum(count[i])
if total == 0:
continue
for j in range(len(count[0])):
count[i][j] /= total
count[i].pop(-1)
total_total = sum([sum(x) for x in count])
for i in range(len(count)):
for j in range(len(count[0])):
count[i][j] /= total_total
return count
def segment_image(img, palette):
img_cpy = img.copy()
h,w,d = img.shape
segments = {}
for col in palette:
segments[col] = []
px2id = [[-1 for j in range(w)] for i in range(h)]
adjacency = {}
visited = set()
seg_id = 0
for i in range(img.shape[0]):
for j in range(img.shape[1]):
if (i,j) in visited:
continue
rgb_hex = get_color(img_cpy,i,j,palette)
#rgb_hex = get_nearest_rgb(img_cpy[i][j],palette)
seg = getsegment(img_cpy,i,j,palette,visited,px2id,adjacency,seg_id)
segments[rgb_hex].append(seg)
seg_id += 1
return (segments, px2id, adjacency)
def preprocess_image(img_num):
testimg_file = os.path.join('test_set2', str(img_num)+'.png')
testimg = Image.open(testimg_file)
testimg = testimg.convert('RGBA')
testimg = np.array(testimg)
with open(os.path.join('test_set2', 'test.csv')) as file:
reader = csv.DictReader(file)
palette = None
for row in reader:
if row['patternId'] == str(img_num):
palette = row['palette'].strip().split(' ')
if palette is None:
print("Bad image ID")
exit(2)
return testimg, palette
def get_color_groups(img_num):
img, palette = preprocess_image(img_num)
segments, px2id, adjacency = segment_image(img, palette)
color_groups = {}
# print(adjacency)
# print(palette)
# img2 = img.copy()
for color in segments:
for seg in segments[color]:
if (15,43) in seg:
print("HELLO!", color)
print(seg)
break
# print(get_color(img2,15,43,palette))
# print(get_color(img2,150,170,palette))
# # print(get_color(img2,3,28,palette))
# for i in range(img2.shape[0]):
# for j in range(img2.shape[1]):
# if px2id[i][j] != 22:# and px2id[i][j] != 36:
# img2[i][j] = [0,0,0,255]
# else:
# print(i,j)
# plt.imshow(img2)
# plt.show()
# print(get_color(img2,1,23,palette))
# print(get_color(img2,1,24,palette))
# print(get_color(img2,3,28,palette))
# for i in range(img2.shape[0]):
# for j in range(img2.shape[1]):
# if px2id[i][j] != 36:# and px2id[i][j] != 36:
# img2[i][j] = [0,0,0,255]
# else:
# print(i,j)
# plt.imshow(img2)
# plt.show()
# print(px2id)
# print(len(adjacency))
enc_str = enclosure_strengths(px2id, len(adjacency), adjacency)
for id1 in adjacency:
for id2 in adjacency[id1]:
#id1 is adjacent to id2
assert(enc_str[id1][id2] > 0)
# adjacency maps from segment id to a set of all segment ids that are adjacent to it
total = sum([sum(x) for x in enc_str])
assert(np.round(total, 2) == 1)
# enc = enc_str[3]
# enc_map = {}
# for i in range(len(enc)):
# if enc[i] != 0:
# enc_map[i] = enc[i]
# print(enc_map)
# print(adjacency[3])
for color in palette:
group = np.full(img.shape, 255)
r,g,b = hex2rgb(color)
for segment in segments[color]:
for px in segment:
group[px[0]][px[1]] = [r,g,b,255]
color_groups[color] = group
return color_groups
def test(img_num):
color_groups = get_color_groups(img_num)
# for color_group in color_groups.values():
# plt.imshow(color_group)
# plt.show()
if __name__ == '__main__':
# test(584317)
test(760602)
# test(465753)