This is a very simple algorithm that is an attempt at contour detection.
I simply look at each pixel and compare it with its 8 surrounding neighbors and sum the the gray scale difference. From there a certain threshold is applied to decide whether to display the pixel or not.
I was expecting something a bit less noisy and more cartoonish but I love the results. It looks as though there's some image compression happening somewhere between my webcam and my pixel array.
This should work pretty good for the video installation I'm working on (projector + infrared webcam + infrared spotlight). See it at BaconWood in May: baconwoodfestival.com/ and hopefully at Burning Man 2010.
For related project, check out my blawg: