forked from robodhruv/vector-valued-regularization
-
Notifications
You must be signed in to change notification settings - Fork 0
/
magnification.m
54 lines (49 loc) · 1.82 KB
/
magnification.m
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
function [new_image, iter] = regularize(img, mask_m, max_iter)
%% Regularization routine modified for magnification
pad_img = padarray(img, [2 2], 0, 'both');
pad_mask = padarray(mask_m, [2 2], 0, 'both');
G = fspecial('gaussian',2, 1);
new_image = pad_img;
[x y] = meshgrid(-2:2, -2:2);
[m n ~] = size(pad_img);
mask = zeros(3,3);
Gx = zeros(m,n,3);
Gy = zeros(m,n,3);
for iter = 1:max_iter
for i = 1:size(img,3)
[Gx(:,:,i),Gy(:,:,i)] = imgradientxy(new_image(:,:,i),'sobel');
end
XX = sum(Gx.^2, 3);
YY = sum(Gy.^2, 3);
XY = sum(Gx.*Gy, 3);
disp(iter);
if (mod(iter, 20) == 0)
imshow(new_image)
end
for i = 3:size(pad_img,1) - 2
for j = 3:size(pad_img,2) - 2
if (pad_mask(i, j) < 1)
continue
end
struct_tensor = [XX(i,j) XY(i,j);XY(i,j) YY(i,j)] ;
struct_tensor = imfilter(struct_tensor, G);
[W,D] = eig(struct_tensor);
[Eigenvalues, permutation] = sort(diag(D), 'descend');
W = W(:, permutation);
T = W(:,1)*W(:,1)'/(1 + Eigenvalues(1) + Eigenvalues(2)) ...
+ W(:,2)*W(:,2)'/sqrt(1+ Eigenvalues(1) + Eigenvalues(2)) ;
T_inv = inv(T);
t = 100;
mask = exp(-((x.^2*T_inv(1,1) + y.^2*T_inv(2,2) + x.*y*(T_inv(1,2) + T_inv(2,1))))/(4*t));
%mask = mask.*(1-pad_mask(i-2:i+2,j-2:j+2));
mask = mask/sum(sum(mask));
mask
for k = 1:size(img,3)
temp_conv = conv2(new_image(i-2:i+2,j-2:j+2,k),mask,'same');
new_image(i,j,k) = temp_conv(3,3);
end
new_image([1,2,size(new_image,1),size(pad_img,1)-1],:) = 0;
new_image(:,[1,2,size(new_image,2),size(pad_img,2)-1]) = 0;
end
end
end