Wow, this one really rocked me back on my heels. Its an interesting way to extract the 'feature codes' that people use in recognition. Going from there into a set of properties that you can feed back into an analysis system would make for a fun machine learning exercise.
Reminds me of "Self-Organizing Maps" [1]<p>1. <a href="https://en.wikipedia.org/wiki/Self-organizing_map" rel="nofollow">https://en.wikipedia.org/wiki/Self-organizing_map</a>