Given a selected pixel, outputs a pixel that is most similar to it.

It takes into consideration all neighboring pixels within a given radius.

Nearest pixel determined by finding the smallest L2 norm distance, using the average of RGB values of all pixels in the given radius.

Here's a sample screenshot. The red square is the selected pixel (it's at the nose).

So I am sure now that initialize the output now with a random texture ... I actually don't even think they pulled the colors even from the source image oddly enough.

ReplyDeleteSo they Find the nearest input neighborhoods then blend input neighborhoods, and just repeating this until "convergence" ... ie either like 40 steps or you get 2 iterations that have the same or very close values