The image that they show reminds me of the way that the 2D discrete cosine transform works. http://openi.nlm.nih.gov/imgs/512/312/2680596/2680596_pone.0... It looks similar to this: https://upload.wikimedia.org/wikipedia/commons/2/23/Dctjpeg....
You'll see these all over the place if you read studies about vision. The (highly oversimplified) reason is that if you imagine the edge-detection processes of a mammalian brain as a set of filters, the impulse response of one of those filters would be a Gabor patch.
This is related to why the DCT is so effective for vision applications, although perhaps less significantly than you might imagine just by looking at the patterns.