There were problems which we encountered implementing the confusing paper presented by Tsai and Huang.  Though our analysis presented for the 1D and 2D case are conceptually correct, there are still some problem in the implementation, and as we have had to make several guesses as to what the authors are referring to in terms of notation, there are several places where problems could be.  We have made every attempt to rectify the situation, and we will present here the best results which we have obtained.

Here two examples from two sets of images that we worked with (the first appears *VERY* dark on this webpage, but when working with it, it is much easier to make out the details):

              .                .

This is what we are able to obtain:

   .

What we notice is that the processed SR image from the first set is indeed very close to being correct.  There are two shifted images which we believe would produce the correct SR image if they would line up correctly.  The second set has a smiliar problem (if looked at more closely), we see that every other pixel contains an erroneous value.