I am probably missing something, but I would think that treating each sub-image as its own grid would be more than enough for a decent mosaic generator. It see

I am probably missing something, but I would think that treating each sub-image as its own grid would be more than enough for a decent mosaic generator. It seems to me you would get your edge detection for free this way. I just don't think it would be hard, but, again, I haven't done it myself so I'm sure there are surprises within.