在Matlab中对图像进行模糊处理-图像尺寸错误

Question

I am trying to do deblurr an image in matlab. this is my code

im = im2double(imread('C:\Users\adhil\Desktop\matlab pics\test.JPG'));
figure, imshow (G1); 
% FFT for B1
G_1 = fftshift(G1); 
G_1 = fft2(G_1); 
G_1 = ifftshift(G_1);

h_1 = fspecial( 'gaussian', [130 221] , 1.0 );
% Fourier Transform of 2D Gaussian 
H_1 = fftshift(h_1);
H_1 = fft2(H_1); 
H_1 = ifftshift(H_1); 

% Apply the filter for Image G_1

display(size(G_1));
display(size(H_1));
F_1a = G_1 ./ H_1; 
F_1a = ifftshift (F_1a); 
F_1a = ifft2 (F_1a); 
F_1a = fftshift (F_1a); 
figure, imshow (F_1a);

However i'm getting the following error

Error using ./
Matrix dimensions must agree.

Error in deblur (line 18)
F_1a = G_1 ./ H_1;

I've noticed that the array dimensions of my image is

display(size(G_1));
ans = 130   221    3
display(size(H_1));
ans = 130   221

However,

h_1 = fspecial( 'gaussian', [130 221  3] , 1.0 );

doesn't take in a 3 dimensional array, please advise

Answer 1

You are basically trying to deblur a colour image but the process you speak of (deconvolution) assumes a grayscale image. You are getting a dimensions mismatch because doing ./ assumes that the matrices you are dividing with are the same dimensions because this is an element-wise operator.

BTW, your beginning code of using fftshift / ifftshift / fft2 for multi-channel images has undefined behaviour for colour images. Specifically, for fftshift/ifftshift , the swapping of the quadrants is done in all dimensions and that will provide behaviour that is undesired when it comes to colour images. Specifically, you probably want to do this for each colour plane separately, and so you'd want to apply a fftshift/ifftshift to each colour plane by itself.

Therefore, one way to simultaneously solve your problem and get rid of this undefined behaviour would be to deblur each channel individually then merge the results. Therefore, wrap all of this in one for loop and define an empty output image to place the results of each channel in their respective positions. Also, your code currently wouldn't run. You define the input image to be im , yet you are using a variable G1 ... which isn't defined. I'm going to assume that this im is actually G1 . The changes I've made to your code have the %// Change comments dispersed throughout:

G1 = im2double(imread('C:\Users\adhil\Desktop\matlab pics\test.JPG')); %// Change
figure, imshow (G1);

h_1 = fspecial( 'gaussian', [130 221] , 1.0 ); %// Change - leave outside loop - never changes
% Fourier Transform of 2D Gaussian 
H_1 = fftshift(h_1);
H_1 = fft2(H_1); 
H_1 = ifftshift(H_1); 

%// Change
F_1a = zeros(size(G1));

for idx = 1 : size(G1, 3) %// Change
    % FFT for B1        
    G_1 = fftshift(G1(:,:,idx)); %// Change
    G_1 = fft2(G_1); 
    G_1 = ifftshift(G_1);

    % Apply the filter for Image G_1
    %// Change
    tmp = G_1 ./ H_1; 
    tmp = ifftshift(tmp); 
    tmp = ifft2(tmp); 
    tmp = fftshift(tmp); 
    F_1a(:,:,idx) = tmp;
end

figure, imshow (F_1a);

Take note that I'm applying a for loop and applying the deblur to each channel individually. I've also taken the Gaussian blur outside of the loop as you only need to take the Fourier Transform once. Once I'm done, I show the results.