Essentially you just flip left to right, using for example fliplr, as you would in the space domain, except that you need to move the zero-frequency column back to the left of the matrix, and also restore the origin back in the space domain.
For example
% get a test image
im = double(imread('pout.tif'))/256;
% move to the frequency domain
fim = fft2(im);
% mirror image in the frequency domain
% circshift one column to the right corrects the origin
fim_mirror = circshift(fliplr(fim), [0 1]);
% back to the space domain to see if it worked
% circshift one column to the left corrects the origin
im_mirror = circshift(ifft2(fim_mirror), [0 -1]);
% compare differences to check result, and display the image
max(max(abs(im - fliplr(im_mirror))))
imshow(im);
figure;
imshow(im_mirror);
If the final one-column shift in the space domain needs to be done in the frequency domain also, you can just multiply by a phase factor.
