-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathcompression.py
186 lines (160 loc) · 7.61 KB
/
compression.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
from skimage import data
import skimage
import skimage.transform as transform
from skimage.color import rgb2gray
import numpy as np
import matplotlib.pyplot as plt
import scipy.ndimage as ndimage
angles = [0, 90, 180, 270]
directions = [1, -1]
def rotation(block, angle):
return transform.rotate(block, angle)
def reflection(block, direction):
return block[::direction, :]
def down_sample(block, factor):
return transform.downscale_local_mean(block, (factor, factor))
def apply_transforms(block, angle, direction):
return rotation(reflection(block, direction), angle)
def apply_contrat_transforms(block, angle, direction, contrast, brightness):
return contrast*rotation(reflection(block, direction), angle) + brightness
def fit_contrast_brightness(block1, block2):
A = np.c_[np.ones(block2.size), block2.reshape(block2.size)]
b = block1.reshape(block1.size)
x, residuals, rank, s = np.linalg.lstsq(A,b)
return x[1], x[0]
def get_transformed_blocks(img, region_size, domaine_size):
transformed_blocks = list()
factor = domaine_size//region_size
for i in range(len(img)//domaine_size + 1):
for j in range(len(img)//domaine_size + 1):
for direction, angle in [(direction, angle) for direction in directions for angle in angles]:
block = img[i*region_size:i*region_size + domaine_size, j*region_size:j*region_size + domaine_size]
transformed_blocks.append([i, j, angle, direction, apply_transforms(down_sample(block, factor), angle, direction)])
return transformed_blocks
def compress(img, region_size, domaine_size, to_csv=False, name="Save.csv"):
transformed_blocks = get_transformed_blocks(img, region_size, domaine_size)
transforms = [[None for k in range(len(img) // region_size)] for j in range(len(img) // region_size)]
c = 0
for i in range(len(img) // region_size):
for j in range(len(img) // region_size):
if c%64 == 0:
print("Step {}/{}".format(c+1, len(img) // region_size * len(img) // region_size))
min_ms = np.inf
for k, l, angle, direction, transformed_block in transformed_blocks:
domaine = transformed_block
region = img[i*region_size:(i+1)*region_size, j*region_size:(j+1)*region_size]
contrast, brightness = fit_contrast_brightness(region, domaine)
ms_diff = np.sum(np.square(contrast*domaine + brightness - region))
if ms_diff < min_ms:
min_ms = ms_diff
i_, j_, angle_, direction_, contrast_, brightness_ = k, l, angle, direction, contrast, brightness
transforms[i][j] = [i_, j_, angle_, direction_, contrast_, brightness_]
c += 1
return transforms
def decompress(transformations, region_size, domaine_size, img_size, nbr_iter, save_every=2, name="name", save_fig=False):
current_image = np.random.rand(img_size[0], img_size[1])
factor = domaine_size // region_size
checkpoints = list()
for it in range(nbr_iter):
if save_fig:
checkpoints.append(current_image)
plt.imshow(current_image, cmap="gray")
plt.savefig("{}{}.png".format(name, it))
print("Step {}/{}".format(it+1, nbr_iter))
for i in range(img_size[0] // region_size):
for j in range(img_size[0] // region_size):
k, l, angle, direction, contrast, brightness = transformations[i][j]
domaine = down_sample(current_image[k*region_size:k*region_size + domaine_size, l*region_size:l*region_size + domaine_size], factor)
block = apply_contrat_transforms(domaine, angle, direction, contrast, brightness)
current_image[i*region_size:(i+1)*region_size, j*region_size:(j+1)*region_size] = block
return current_image
def test_compress():
#img = rgb2gray(data.astronaut())
img = skimage.io.imread("monkey.gif", as_grey=True)
img = down_sample(img, 2)
plt.imshow(img, cmap="gray")
plt.show()
print(img.shape)
b = compress(img, 4, 8)
print("-- Image compressed")
img_dec = decompress(b, 4, 8, img.shape, 8, name="astronaut", save_fig=True)
print("-- Image decompressed")
plt.imshow(img_dec, cmap="gray")
plt.show()
return img_dec
def compute_PSNR(nbr_step):
ref = down_sample(skimage.io.imread("monkey.gif", as_grey=True), 4)
psnr = list()
init_region = 2
init_domaine = 64
for k in range(nbr_step):
print("Compressing image with {} domaine size - {} region size".format(init_domaine//(2**k), init_region))
transforms = compress(ref, init_region, init_domaine//(2**k))
print("Decompressing image")
img_dec = decompress(transforms, init_region, init_domaine//(2**k), ref.shape, 8)
EQM_ = np.mean(np.sum(np.square(ref - img_dec)))
psnr_ = (10/np.log(10)) * np.log(255**2/EQM_)
psnr.append(psnr_)
psnr_2 = list()
init_region = 8
for k in range(nbr_step):
print("Compressing image with {} domaine size - {} region size".format(init_domaine//(2**2), init_region//(2**k)))
transforms = compress(ref, init_region//(2**k), init_domaine//(2**2))
print("Decompressing image")
img_dec = decompress(transforms, init_region//(2**k), init_domaine//(2**2), ref.shape, 8)
EQM_ = np.mean(np.sum(np.square(ref - img_dec)))
psnr_2_ = (10/np.log(10)) * np.log(255**2/EQM_)
psnr_2.append(psnr_2_)
plt.figure()
plt.plot([init_domaine//(2**k) for k in range(nbr_step)], psnr, marker='x', linestyle="--", label="psnr")
plt.title("PSNR vs Domain Size")
plt.xlabel("Domain Size")
plt.ylabel("PSNR")
plt.legend()
plt.plot([init_domaine//(2**k) for k in range(nbr_step)], psnr_2, marker='x', linestyle="--", label="psnr", color="r")
plt.savefig("error_rate_3.png")
plt.show()
return psnr
def compute_SSIM(nbr_step):
ref = down_sample(skimage.io.imread("monkey.gif", as_grey=True), 4)
print(ref.shape)
SSIM = list()
init_region = 4
init_domaine = 64
c1 = (0.01*255)**2
c2 = (0.03*255)**2
c3 = c2/2
for k in range(nbr_step):
print("Compressing image with {} domaine size - {} region size".format(init_domaine//(2**k), init_region))
transforms = compress(ref, init_region, init_domaine//(2**k))
print("Decompressing image")
img_dec = decompress(transforms, init_region, init_domaine//(2**k), ref.shape, 8)
print(img_dec.shape)
SSIM_ = list()
for l in range(img_dec.shape[0] // 8):
for j in range(img_dec.shape[0] // 8):
x, y = ref[8*k:8*(k+1), 8*j:8*(j+1)], img_dec[8*k:8*(k+1), 8*j:8*(j+1)]
mux = np.mean(x)
muy = np.mean(y)
sigx = np.std(x)
sigy = np.std(y)
cov = np.cov(x, y)
l = (2*mux*muy + c1)/(mux**2 + muy**2 + c1)
c = (2*sigx*sigy + c2)/(sigx**2 + sigy**2 + c2)
s = (cov + c3)/(sigx*sigy + c3)
SSIM_.append(l*c*s)
SSIM.append(np.mean(SSIM_))
print(len(SSIM_))
print(np.mean(SSIM_))
plt.imshow(img_dec, cmap="gray", interpolation=None)
plt.savefig("monkey{}-{}.png".format(init_domaine//(2**k), init_region))
plt.figure()
plt.plot([init_domaine//(2**k) for k in range(nbr_step)], SSIM, marker='x', linestyle="--", label="SSIM")
plt.xticks([init_domaine//(2**k) for k in range(nbr_step)])
plt.title("SSIM vs Domain Size")
plt.xlabel("Domain Size")
plt.ylabel("SSIM")
plt.legend()
plt.savefig("error_SSIM5.png")
plt.show()
compute_SSIM(4)