
If you want to use some other images, you will need to prewhiten them
first.  If the images are NxN pixels, then you will need to multiply
by the filter f*exp(-(f/f_0)^4) in the frequency domain, where
f_0=0.4*N.  Once you have preprocessed a number of images this way,
all the same size, then you should reshape each image into a column
vector and concatenate these columns into a N^2 x M matrix, where M is
the number of images.  Then rescale this array so that the average
image variance is 0.1.  Name this matrix IMAGES, save it to a file for
future use, and you should be off and running.  The following commands
should do this:

N=images_size;
M=num_images;

[fx fy]=meshgrid(-N/2:N/2-1,-N/2:N/2-1);
rho=sqrt(fx.*fx+fy.*fy);
f_0=0.4*N;
filt=rho.*exp(-(rho/f_0).^4);

for i=1:M
  image=get_image;
  If=fft2(image);
  imagew=real(ifft2(If.*fftshift(filt)));
  IMAGES(:,i)=reshape(imagew,N^2,1);
end

IMAGES=sqrt(0.1)*IMAGES/sqrt(mean(var(IMAGES)));

save IMAGES IMAGES

