-
Notifications
You must be signed in to change notification settings - Fork 9
/
cross_correlation.py
154 lines (117 loc) · 6.05 KB
/
cross_correlation.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
from math import ceil
import numpy as np
import torch
import torch.nn.functional as F
from skimage.transform.integral import integral_image as integral
def xcorr(img_1, img_2, half_width):
"""
Cross-Correlation Field computation.
Parameters
----------
img_1 : Numpy Array
First image on which calculate the cross-correlation. Dimensions: H, W
img_2 : Numpy Array
Second image on which calculate the cross-correlation. Dimensions: H, W
half_width : int
The semi-size of the window on which calculate the cross-correlation
Return
------
L : Numpy array
The cross-correlation map between img_1 and img_2
"""
w = ceil(half_width)
ep = 1e-20
if (len(img_1.shape)) != 3:
img_1 = np.expand_dims(img_1, axis=-1)
if (len(img_2.shape)) != 3:
img_2 = np.expand_dims(img_2, axis=-1)
img_1 = np.pad(img_1.astype(np.float64), ((w, w), (w, w), (0, 0)))
img_2 = np.pad(img_2.astype(np.float64), ((w, w), (w, w), (0, 0)))
img_1_cum = np.zeros(img_1.shape)
img_2_cum = np.zeros(img_2.shape)
for i in range(img_1.shape[-1]):
img_1_cum[:, :, i] = integral(img_1[:, :, i]).astype(np.float64)
for i in range(img_2.shape[-1]):
img_2_cum[:, :, i] = integral(img_2[:, :, i]).astype(np.float64)
img_1_mu = (img_1_cum[2 * w:, 2 * w:, :] - img_1_cum[:-2 * w, 2 * w:, :] - img_1_cum[2 * w:, :-2 * w,
:] + img_1_cum[:-2 * w, :-2 * w, :]) / (
4 * w ** 2)
img_2_mu = (img_2_cum[2 * w:, 2 * w:, :] - img_2_cum[:-2 * w, 2 * w:, :] - img_2_cum[2 * w:, :-2 * w,
:] + img_2_cum[:-2 * w, :-2 * w, :]) / (
4 * w ** 2)
img_1 = img_1[w:-w, w:-w, :] - img_1_mu
img_2 = img_2[w:-w, w:-w, :] - img_2_mu
img_1 = np.pad(img_1.astype(np.float64), ((w, w), (w, w), (0, 0)))
img_2 = np.pad(img_2.astype(np.float64), ((w, w), (w, w), (0, 0)))
i2 = img_1 ** 2
j2 = img_2 ** 2
ij = img_1 * img_2
i2_cum = np.zeros(i2.shape)
j2_cum = np.zeros(j2.shape)
ij_cum = np.zeros(ij.shape)
for i in range(i2_cum.shape[-1]):
i2_cum[:, :, i] = integral(i2[:, :, i]).astype(np.float64)
for i in range(j2_cum.shape[-1]):
j2_cum[:, :, i] = integral(j2[:, :, i]).astype(np.float64)
for i in range(ij_cum.shape[-1]):
ij_cum[:, :, i] = integral(ij[:, :, i]).astype(np.float64)
sig2_ij_tot = (ij_cum[2 * w:, 2 * w:, :] - ij_cum[:-2 * w, 2 * w:, :] - ij_cum[2 * w:, :-2 * w, :] + ij_cum[:-2 * w,
:-2 * w, :])
sig2_ii_tot = (i2_cum[2 * w:, 2 * w:, :] - i2_cum[:-2 * w, 2 * w:, :] - i2_cum[2 * w:, :-2 * w, :] + i2_cum[:-2 * w,
:-2 * w, :])
sig2_jj_tot = (j2_cum[2 * w:, 2 * w:, :] - j2_cum[:-2 * w, 2 * w:, :] - j2_cum[2 * w:, :-2 * w, :] + j2_cum[:-2 * w,
:-2 * w, :])
# sig2_ij_tot = np.clip(sig2_ij_tot, ep, sig2_ij_tot.max())
sig2_ii_tot = np.clip(sig2_ii_tot, ep, sig2_ii_tot.max())
sig2_jj_tot = np.clip(sig2_jj_tot, ep, sig2_jj_tot.max())
L = sig2_ij_tot / ((sig2_ii_tot * sig2_jj_tot) ** 0.5 + ep)
return L
def xcorr_torch(img_1, img_2, half_width, device):
"""
A PyTorch implementation of Cross-Correlation Field computation.
Parameters
----------
img_1 : Torch Tensor
First image on which calculate the cross-correlation. Dimensions: 1, 1, H, W
img_2 : Torch Tensor
Second image on which calculate the cross-correlation. Dimensions: 1, 1, H, W
half_width : int
The semi-size of the window on which calculate the cross-correlation
device : Torch Device
The device on which perform the operation.
Return
------
L : Torch Tensor
The cross-correlation map between img_1 and img_2
"""
w = ceil(half_width)
ep = 1e-20
img_1 = img_1.type(torch.DoubleTensor)
img_2 = img_2.type(torch.DoubleTensor)
img_1 = img_1.to(device)
img_2 = img_2.to(device)
img_1 = F.pad(img_1, (w, w, w, w))
img_2 = F.pad(img_2, (w, w, w, w))
img_1_cum = torch.cumsum(torch.cumsum(img_1, dim=-1), dim=-2)
img_2_cum = torch.cumsum(torch.cumsum(img_2, dim=-1), dim=-2)
img_1_mu = (img_1_cum[:, :, 2 * w:, 2 * w:] - img_1_cum[:, :, :-2 * w, 2 * w:] - img_1_cum[:, :, 2 * w:, :-2 * w] +
img_1_cum[:, :, :-2 * w, :-2 * w]) / (4 * w ** 2)
img_2_mu = (img_2_cum[:, :, 2 * w:, 2 * w:] - img_2_cum[:, :, :-2 * w, 2 * w:] - img_2_cum[:, :, 2 * w:, :-2 * w] +
img_2_cum[:, :, :-2 * w, :-2 * w]) / (4 * w ** 2)
img_1 = img_1[:, :, w:-w, w:-w] - img_1_mu
img_2 = img_2[:, :, w:-w, w:-w] - img_2_mu
img_1 = F.pad(img_1, (w, w, w, w))
img_2 = F.pad(img_2, (w, w, w, w))
i2_cum = torch.cumsum(torch.cumsum(img_1 ** 2, dim=-1), dim=-2)
j2_cum = torch.cumsum(torch.cumsum(img_2 ** 2, dim=-1), dim=-2)
ij_cum = torch.cumsum(torch.cumsum(img_1 * img_2, dim=-1), dim=-2)
sig2_ij_tot = (ij_cum[:, :, 2 * w:, 2 * w:] - ij_cum[:, :, :-2 * w, 2 * w:] - ij_cum[:, :, 2 * w:, :-2 * w] +
ij_cum[:, :, :-2 * w, :-2 * w])
sig2_ii_tot = (i2_cum[:, :, 2 * w:, 2 * w:] - i2_cum[:, :, :-2 * w, 2 * w:] - i2_cum[:, :, 2 * w:, :-2 * w] +
i2_cum[:, :, :-2 * w, :-2 * w])
sig2_jj_tot = (j2_cum[:, :, 2 * w:, 2 * w:] - j2_cum[:, :, :-2 * w, 2 * w:] - j2_cum[:, :, 2 * w:, :-2 * w] +
j2_cum[:, :, :-2 * w, :-2 * w])
sig2_ii_tot = torch.clip(sig2_ii_tot, ep, sig2_ii_tot.max().item())
sig2_jj_tot = torch.clip(sig2_jj_tot, ep, sig2_jj_tot.max().item())
L = sig2_ij_tot / ((sig2_ii_tot * sig2_jj_tot) ** 0.5 + ep)
return L