-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathwatermark_remover.py
302 lines (245 loc) · 9.06 KB
/
watermark_remover.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
import cv2 as cv
import numpy as np
import time
import os
def get_alpha_W_white():
video = "videos/stock-white.mp4"
cap = cv.VideoCapture(video)
res, waterpic = cap.read()
# cv.imshow("waterpic", waterpic)
waterpic = waterpic.min(axis=-1)
# print(np.bincount(waterpic.reshape(-1), minlength=256))
waterpic = waterpic.astype(np.float)
waterpic[waterpic>247] = 0
waterpic[waterpic!=0] = 223.5
# cv.imshow("waterpic2", waterpic.astype(np.uint8))
alpha_0 = 32/255
alpha = np.ones(waterpic.shape) * alpha_0
alpha[waterpic==0] = 0
J = waterpic
W = np.zeros(waterpic.shape, dtype=np.float)
mask = waterpic!=0
W[mask] = (J[mask] - (1 - alpha[mask])*255)/alpha[mask]
# print(J[mask][0], (1 - alpha[mask][0])*255)
# print(W[W<0])
# print(W[W>255])
# print(W[W!=0][0])
# print(alpha[mask][0]*W[mask][0])
# cv.imshow("white alpha,W",np.vstack([alpha*255, W]).astype(np.uint8))
# test_img1 = ((1-alpha)*np.zeros(W.shape) + alpha*W).astype(np.uint8)
# test_img2 = ((1-alpha)*np.ones(W.shape)*255 + alpha*W).astype(np.uint8)
# cv.imshow("test_img", np.vstack([test_img1,test_img2]))
# cv.waitKey(0)
alpha = np.repeat(alpha, 3, axis=-1).reshape(*(alpha.shape),3)
W = np.repeat(W, 3, axis=-1).reshape(*(W.shape),3)
return alpha, W
def get_alpha_W_black():
video = "videos/stock-black.mp4"
cap = cv.VideoCapture(video)
res, waterpic = cap.read()
waterpic = waterpic.max(axis=-1)
# print(np.bincount(waterpic.reshape(-1),minlength=256))
# cv.imshow("waterpic", waterpic)
alpha_0 = 31.5/255
thresh = 5
waterpic[waterpic<thresh] = 0
waterpic[waterpic>=thresh] = 31.5
alpha = np.ones(waterpic.shape) * alpha_0
alpha[waterpic==0] = 0
W = waterpic/alpha_0
# cv.imshow("black alpha,W",np.vstack([alpha*255, W]).astype(np.uint8))
# test_img1 = ((1-alpha)*np.zeros(W.shape) + alpha*W).astype(np.uint8)
# test_img2 = ((1-alpha)*np.ones(W.shape)*255 + alpha*W).astype(np.uint8)
# cv.imshow("test_img", np.vstack([test_img1,test_img2]))
# cv.waitKey(0)
alpha = np.repeat(alpha, 3, axis=-1).reshape(*(alpha.shape),3)
W = np.repeat(W, 3, axis=-1).reshape(*(W.shape),3)
return alpha, W
def move_box(alpha, W, offset):
pos = alpha.nonzero()
box = [pos[0].min(), pos[1].min(), pos[0].max(), pos[1].max()]
new_alpha = np.zeros(alpha.shape, dtype=alpha.dtype)
new_W = np.zeros(alpha.shape, dtype=W.dtype)
box_2 = [box[i]-offset[i%2] for i in range(4)]
idx1 = np.ix_(range(box[0],box[2]), range(box[1], box[3]))
idx2 = np.ix_(range(box_2[0],box_2[2]), range(box_2[1], box_2[3]))
new_alpha[idx2] = alpha[idx1]
new_W[idx2] = W[idx1]
return new_alpha, new_W
def merge_W(alpha1,W1,alpha2,W2):
# alpha1 是外轮廓，外接矩形比较大
alpha1, W1 = move_box(alpha1, W1, [0,-2])
alpha2, W2 = move_box(alpha2, W2, [4,0])
alpha = alpha1 + alpha2
alpha[alpha>alpha2.max()] = alpha2.max()
W = W1 + W2
W[W>W1.max()] = W1.max()
# cv.imshow("alpha,W",np.vstack([alpha*255, W]).astype(np.uint8))
# cv.waitKey(0)
return alpha, W
def find_offset(img1, img2):
# 找到img1中的水印需要移动多少才能刚好与img2重合
img1 = cv.cvtColor(img1.astype(np.uint8), cv.COLOR_BGR2GRAY)
pos = img1.nonzero()
box = [pos[0].min(), pos[1].min(), pos[0].max(), pos[1].max()]
idx1 = np.ix_(range(box[0],box[2]), range(box[1], box[3]))
img1 = img1[idx1]
# cv.imshow("img1",img1.astype(np.uint8))
box2 = [box[0]-10, box[1]-10, box[2]+10, box[3]+10]
idx2 = np.ix_(range(box2[0],box2[2]), range(box2[1], box2[3]))
img2 = img2[idx2]
img2 = cv.cvtColor(img2.astype(np.uint8), cv.COLOR_BGR2GRAY)
# cv.imshow("img2",img2.astype(np.uint8))
# img3 = cv.filter2D(img2.astype(np.float), -1, img1.astype(np.float))
img3 = cv.matchTemplate(img2,img1,cv.TM_CCORR_NORMED)
# print(img1.shape,img2.shape,img3.shape)
cent = [img3.shape[0]//2, img3.shape[1]//2]
img3 = img3[cent[0]-10:cent[0]+10, cent[1]-10:cent[1]+10]
maxindex = img3.argmax()
row, col = maxindex//img3.shape[1], maxindex%img3.shape[1]
print((row,col))
assert img3[row,col]==img3.max()
# cv.imshow("f2d", img3/img3.max())
# cv.waitKey(0)
return [10-row, 10-col]
def blur_mask(alpha, W):
alpha,W = move_box(alpha, W, [-4,0])
video = "videos/stock-white.mp4"
video = "/database/水印视频/shutterstock/多旋翼无人机/stock-footage-drone-landing-in-hand.mp4"
cap = cv.VideoCapture(video)
pos = alpha.max(axis=-1).nonzero()
box = [pos[0].min(), pos[1].min(), pos[0].max(), pos[1].max()]
difs = []
if False:
index = 0
while True:
result, frame = cap.read()
if not result: break
index += 1
I = (frame.astype(float) - alpha * W)/(1-alpha)
# cv.imshow("sub result",(frame.astype(float) - alpha * W).astype(np.uint8))
# print(frame[alpha!=0].min(), frame[alpha!=0].max(), )
# print(I.min(),I.max(),alpha.shape,len(I[I<0]), np.mean(I[I<0]), len(I[I>255]), np.mean(I[I>255]))
# print(frame[I>255][0:2], alpha[I>255][0:2], W[I>255][0:2], I[I>255][0:2])
I[I>255]=255
I[I<0] = 0
I = I.astype(np.uint8)
fI = I.copy()
fI[box[0]:box[2],box[1]:box[3],:] = cv.medianBlur(fI[box[0]:box[2],box[1]:box[3],:], 5)
oboximage = frame[box[0]:box[2],box[1]:box[3],:]
iboximage = I[box[0]:box[2],box[1]:box[3],:]
fboximage = fI[box[0]:box[2],box[1]:box[3],:]
dif = np.abs(fboximage.astype(float)-iboximage.astype(float))
difs.append(dif)
# cv.rectangle(I,(box[1],box[0]),(box[3],box[2]),(0, 255, 255),1,8)
# cv.imshow("diffenence", np.vstack([oboximage,iboximage,fboximage,
# dif.astype(np.uint8),np.median(np.array(difs),axis=0).astype(np.uint8)]))
# cv.waitKey(0)
# if index==100:break
np.save("med.npy", np.array(difs))
difs = np.load("med.npy")
difs = difs.astype(np.uint8)
if False:
med = np.zeros(difs[0].shape[0:2], dtype=np.uint8)
for i in range(med.shape[0]):
for j in range(med.shape[1]):
med[i,j] = np.bincount(difs[:,i,j,:].reshape(-1),minlength=256).argmax()
else:
med = np.median(difs,axis=0).astype(np.uint8)
med = med.max(axis=-1)
med[med<3]=0
print(med.shape)
med = np.repeat(med, 3, axis=-1).reshape(*(med.shape),3)
cv.imwrite("imgs/shutterstock_difference.png",med)
# cv.imshow("dif", med)
bg = np.zeros(W.shape)
bg[box[0]:box[2],box[1]:box[3],:] = med
# bg,_ = move_box(bg,bg,[0,-2])
# cv.imshow("bg", bg.astype(np.uint8)+(alpha*100).astype(np.uint8))
# cv.waitKey(0)
return bg
def process_video(video, alpha, W, bg, out_video=None):
cap = cv.VideoCapture(video)
if out_video is not None:
video_FourCC= int(cap.get(cv.CAP_PROP_FOURCC))
video_fps = cap.get(cv.CAP_PROP_FPS)
video_size = (int(cap.get(cv.CAP_PROP_FRAME_WIDTH)),
int(cap.get(cv.CAP_PROP_FRAME_HEIGHT)))
vwriter = cv.VideoWriter(out_video, video_FourCC,
video_fps, video_size, isColor=True)
else:
select = set(range(1,1000,80))
accum_time = 0
index = 0
while True:
res, frame = cap.read()
if not res:break
start = time.time()
if frame.shape!=(336,596,3):
break
index += 1
if index==1:
offset = find_offset(W, frame)
print(offset)
# offset = [-4,-2]
alpha, W = move_box(alpha, W, offset)
bg, _ = move_box(bg, bg, offset)
pos = alpha.nonzero()
box = [pos[0].min(), pos[1].min(), pos[0].max(), pos[1].max()]
ROI = np.ix_(range(box[0]-3,box[2]+3), range(box[1]-3,box[3]+3))
aW = (alpha*W)[ROI]
a1 = (1/(1 - alpha))[ROI]
med = bg[ROI]
# print(aW.shape)
if out_video is None and index not in select:continue
org_frame = frame.copy()
frame = frame.astype(np.float)
J = frame[ROI]
I = (J - aW) * a1
I[I<0] = 0
I[I>255] = 255
frame[ROI] = I
sub_frame = frame.copy().astype(np.uint8)
org_fI = I.copy()
fI = I.copy()
fI = cv.medianBlur(fI.astype(np.uint8), 5)
I[med!=0] = fI[med!=0]
I[I<0] = 0
I[I>255] = 255
frame[ROI] = I
frame = frame.astype(np.uint8)
accum_time += time.time() - start
print("frame:", index,"fps:", index/accum_time)
# cv.imwrite("imgs/shutterstock_median_blur.png",fI)
# cv.imwrite("imgs/shutterstock_median_subs.png",org_fI.astype(np.uint8))
# cv.imwrite("imgs/shutterstock_median_res.png",I.astype(np.uint8))
# cv.imwrite("imgs/shutterstock_orig.png",org_frame)
# cv.imwrite("imgs/shutterstock_subs.png",sub_frame)
# cv.imwrite("imgs/shutterstock_res.png",frame)
cv.imshow("frame", np.vstack([org_frame,frame]))
cv.waitKey(0)
if out_video:vwriter.write(frame)
cap.release()
if out_video:vwriter.release()
def main():
alpha1, W1 = get_alpha_W_black()
alpha2, W2 = get_alpha_W_white()
alpha, W = merge_W(alpha1,W1,alpha2,W2)
bg = blur_mask(alpha, W)
# cv.imwrite("imgs/shutterstock_alpha.png", (alpha*255).astype(np.uint8))
# cv.imwrite("imgs/shutterstock_W.png", (W).astype(np.uint8))
video = "/database/水印视频/shutterstock/多旋翼无人机/stock-footage-drone-landing-in-hand.mp4"
process_video(video, alpha, W, bg)
return
dirname = "/database/水印视频/shutterstock/"
for root, dirs, names in os.walk(dirname):
if dirs==[]:
os.makedirs(root.replace("水印视频", "去水印视频"), exist_ok=True)
for name in names:
video = root+"/"+name
out_video = root.replace("水印视频", "去水印视频")+"/"+name
print(video)
if not os.path.exists(out_video):
process_video(video, alpha, W, bg, out_video)
if __name__ == '__main__':
main()