From CS294-69 Image Manipulation and Computational Photography Fa11
These are two original images that I use:
These are the fft transforms of those images
This is the hybrid image obtained by mixing the two filtered images. The image looks different from different distances.
I apply a convolution of a 3 by 3 moving average matrix (lowpass filter) to the hybrid image to filter out high frequencies. The lowpass-filtered images is then subtracted from the hybrid image to obtain the prediction error L0. We can see that only features of the high-frequency image are retained as the process acts as a highpass filter.
The filtered hybrid image is then reduced by half on each dimension and applied the same 3 by 3 lowpass filter.
The process is continued ...