Austin (austin_biceps)

When analyzing images, constraint propagation works because relevant surface properties change slowly almost everywhere. Surface distance and surface direction can change rapidly at edges, of course, but edge points represent only a relatively small number of the points in an image array.