-
Notifications
You must be signed in to change notification settings - Fork 3
/
main.py
225 lines (203 loc) · 8.36 KB
/
main.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
import numpy as np
import cv2
import copy
import cv_joke as cv_jk
import scipy.signal as sci
import matrix as mt
import random
import time
'''将默认值为0的内核的某一区域全置为1'''
def ones_kernel(kernel,size=(1,1),loc=(0,0),value = 1):
tmp = np.ones(size)
kernel_tmp = copy.deepcopy(kernel)
kernel_tmp[loc[0]:(loc[0]+size[0]),loc[1]:(loc[1]+size[1])] = tmp
return kernel_tmp
'''Side Mean Filter实现'''
def s_meanfilter(img,radius,iteration = 1):
r = radius
zero_kernel = np.zeros([2*r+1,2*r+1])
k_L = ones_kernel(zero_kernel,size= (2*r+1,r+1),loc= (0,0))/((2*r+1)*(r+1))
k_R = ones_kernel(zero_kernel,size= (2*r+1,r+1),loc= (0,r))/((2*r+1)*(r+1))
K_U = k_L.T
k_D = K_U[::-1]
k_NW = ones_kernel(zero_kernel,size= (r+1,r+1),loc= (0,0))/((r+1)*(r+1))
k_NE = ones_kernel(zero_kernel,size= (r+1,r+1),loc= (0,r))/((r+1)*(r+1))
k_SW = k_NW[::-1]
k_SE = k_NE[::-1]
kernels = [k_L,k_R,K_U,k_D,k_NW,k_NE,k_SW,k_SE]
m = img.shape[0]+2*r
n = img.shape[1]+2*r
dis = np.zeros([8,m,n]);
result = copy.deepcopy(img)
for ch in range(img.shape[2]):
U = np.pad(img[:,:,ch],(r,r),'edge');
for i in range(iteration):
for id,kernel in enumerate(kernels):
conv2 = sci.correlate2d(U,kernel,'same')
dis[id] = conv2 - U
U = U + mt.mat_absmin(dis)
result[:,:,ch] = U[r:-r,r:-r]
return result
'''滑动内核与矩阵星乘取中值'''
def mid_mult(img,kernel,r,start_offset,end_offset):
result = []
for row in range(start_offset[0],img.shape[0]-kernel.shape[0]+1-end_offset[0]):
for col in range(start_offset[1],img.shape[1]-kernel.shape[1]+1-end_offset[1]):
#img_roi = copy.deepcopy(img[row:row+kernel.shape[0],col:col+kernel.shape[1]])
img_roi = img[row:row+kernel.shape[0],col:col+kernel.shape[1]]
mid_tmp = np.median(img_roi*kernel)
result.append(mid_tmp)
result = np.reshape(np.array(result),(img.shape[0]-2*r,img.shape[1]-2*r))
result = np.pad(result,(r,r),'edge');
return result
'''Side Median Filter实现'''
def s_medianfilter(img,radius,iteration = 1):
r = radius
# 异型内核
k_L = np.ones((2*r+1,r+1))
k_R = k_L
k_U = k_L.T
k_D = k_U
k_NW = np.ones((r+1,r+1))
k_NE = k_NW
k_SW = k_NW
k_SE = k_NW
kernels = [k_L,k_R,k_U,k_D,k_NW,k_NE,k_SW,k_SE]
start_offsets = [(0,0),(0,r),(0,0),(r,0),(0,0),(0,r),(r,0),(r,r)]
end_offsets = [(0,r),(0,0),(r,0),(0,0),(r,r),(r,0),(0,r),(0,0)]
m = img.shape[0]+2*r
n = img.shape[1]+2*r
dis = np.zeros([8,m,n]);
result = copy.deepcopy(img)
for ch in range(img.shape[2]):
U = np.pad(img[:,:,ch],(r,r),'edge');
for i in range(iteration):
for id in range(len(kernels)):
mid_result = mid_mult(U,kernels[id],r,start_offsets[id],end_offsets[id])
dis[id] = mid_result - U
U = U + mt.mat_absmin(dis)
result[:,:,ch] = U[r:-r,r:-r]
return result
'''将内核的某一区域保留,其余置0'''
def zeros_kernel(kernel,size=(1,1),loc=(0,0)):
kernel_tmp = np.zeros(kernel.shape)
kernel_tmp[loc[0]:(loc[0]+size[0]),loc[1]:(loc[1]+size[1])] = kernel[loc[0]:(loc[0]+size[0]),loc[1]:(loc[1]+size[1])]
kernel_tmp = kernel_tmp/np.sum(kernel_tmp)
return kernel_tmp
'''Side Gaussian Filter实现'''
def s_gausfilter(img,radius,sigma = 0,iteration = 1):
r = radius
gaus_kernel = cv2.getGaussianKernel(2*r+1,sigma) # sigma = ((n-1)*0.5 - 1)*0.3 + 0.8
gaus_kernel = gaus_kernel.dot(gaus_kernel.T)
gaus_kernel = gaus_kernel.astype(np.float)
k_L = zeros_kernel(gaus_kernel,size= (2*r+1,r+1),loc= (0,0))
k_R = zeros_kernel(gaus_kernel,size= (2*r+1,r+1),loc= (0,r))
K_U = k_L.T
k_D = K_U[::-1]
k_NW = zeros_kernel(gaus_kernel,size= (r+1,r+1),loc= (0,0))
k_NE = zeros_kernel(gaus_kernel,size= (r+1,r+1),loc= (0,r))
k_SW = k_NW[::-1]
k_SE = k_NE[::-1]
kernels = [k_L,k_R,K_U,k_D,k_NW,k_NE,k_SW,k_SE]
m = img.shape[0]+2*r
n = img.shape[1]+2*r
dis = np.zeros([8,m,n]);
result = copy.deepcopy(img)
for ch in range(img.shape[2]):
U = np.pad(img[:,:,ch],(r,r),'edge');
for i in range(iteration):
for id,kernel in enumerate(kernels):
conv2 = sci.correlate2d(U,kernel,'same')
dis[id] = conv2 - U
U = U + mt.mat_absmin(dis)
result[:,:,ch] = U[r:-r,r:-r]
return result
'''测试1'''
# image = cv_jk.imread('image\cat.jpg')
# resizeimg = cv_jk.resize(image,(0.8,0.8))
# noise_img = cv_jk.sp_noise(resizeimg,0.02)
# # 均值滤波
# img_mean = cv2.blur(noise_img, (7,7))
# img_mean_5 = noise_img
# img_mean_10 = noise_img
# start = time.process_time()
# for i in range(5):
# img_mean_5 = cv2.blur(img_mean_5, (7,7))
# for i in range(10):
# img_mean_10 = cv2.blur(img_mean_10, (7,7))
# print((time.process_time()-start)/16)
# start = time.process_time()
# # S-均值滤波
# result = s_meanfilter(noise_img,3,1)
# result_5 = s_meanfilter(noise_img,3,5)
# result_10 = s_meanfilter(noise_img,3,10)
# print((time.process_time()-start)/16)
# # cv_jk.imshow('noise_img',noise_img)
# # cv_jk.imshow('img_mean',img_mean)
# # cv_jk.imshow('result',result)
# cv_jk.save_images(resizeimg,r'.\result','source_img','jpg')
# cv_jk.save_images(noise_img,r'.\result','noise_img','jpg')
# cv_jk.save_images(img_mean,r'.\result','img_mean','jpg')
# cv_jk.save_images(img_mean_5,r'.\result','img_mean_5','jpg')
# cv_jk.save_images(img_mean_10,r'.\result','img_mean_10','jpg')
# cv_jk.save_images(result,r'.\result','result','jpg')
# cv_jk.save_images(result_5,r'.\result','result_5','jpg')
# cv_jk.save_images(result_10,r'.\result','result_10','jpg')
# cv2.waitKey(0)
# '''测试2'''
# image = cv_jk.imread('image\cat.jpg')
# resizeimg = cv_jk.resize(image,(0.8,0.8))
# noise_img = cv_jk.sp_noise(resizeimg,0.02)
# # 中值滤波
# img_median = cv2.medianBlur(noise_img, 7)
# img_median_5 = noise_img
# img_median_10 = noise_img
# start = time.process_time()
# for i in range(5):
# img_median_5 = cv2.medianBlur(img_median_5, 7)
# for i in range(10):
# img_median_10 = cv2.medianBlur(img_median_10, 7)
# print((time.process_time()-start)/16)
# start = time.process_time()
# # S-中值滤波
# result = s_medianfilter(noise_img,3,1)
# print('1')
# result_5 = s_medianfilter(noise_img,3,5)
# print('5')
# result_10 = s_medianfilter(noise_img,3,10)
# print('10')
# print((time.process_time()-start)/16)
# # cv_jk.imshow('noise_img',noise_img)
# # cv_jk.imshow('img_median',img_median)
# # cv_jk.imshow('result',result)
# # cv_jk.save_images(resizeimg,r'.\result','source_img','jpg')
# # cv_jk.save_images(noise_img,r'.\result','noise_img','jpg')
# cv_jk.save_images(img_median,r'.\result','trad_median','jpg')
# cv_jk.save_images(img_median_5,r'.\result','trad_median_5','jpg')
# cv_jk.save_images(img_median_10,r'.\result','trad_median_10','jpg')
# cv_jk.save_images(result,r'.\result','s-median_result','jpg')
# cv_jk.save_images(result_5,r'.\result','s-median_result_5','jpg')
# cv_jk.save_images(result_10,r'.\result','s-median_result_10','jpg')
'''测试3'''
image = cv_jk.imread('image\moon.jpg')
resizeimg = cv_jk.resize(image,(0.3,0.3))
noise_img = cv_jk.sp_noise(resizeimg,0.02)
# 高斯滤波
img_gaussion = cv2.GaussianBlur(noise_img,(15,15),0.01)
# S-高斯滤波
result = s_gausfilter(noise_img,7)
cv_jk.imshow('source_img',noise_img)
cv_jk.imshow('img_gaussion',img_gaussion)
cv_jk.imshow('result',result)
cv2.waitKey(0)
'''调试'''
# img_array = np.array([[[0],[1],[2]],[[0],[1],[2]],[[0],[1],[2]]])
# s_gausfilter(img_array,1)
# img_array = np.array([[[0,0,0],[1,1,1]],[[0,0,0],[1,1,1]]])
# result = s_boxfilter(img_array,1,1)
# Hello~你好请问可以请教一下关于您发表的边窗滤波的一点问题吗?
# 就是我用 python 复现了一下文中提到的 S-BOX filter, 虽然最后出来的结果还可以, 但是运行速度很慢
# 在这个过程中其实我怀疑我对这个算法的思想还是拿捏的不准确, 所以想请教你一下~
# 我的理解是生成的 Kernel 大小是(2r+1,2r+1),采取不同方向的时候, 需要对Kernel内容进行调整
# 比如 k_L 就需要将右半部分置0,剩下的左半部分内核取均值，而 k_NW 这样斜向的就只保留角落的一块区域并取均值。
# 然后用这8个(2r+1,2r+1)的核去与图像作卷积，取 dis 最小的卷积结果作为最后该像素点的像素值