Yes I'm using a similar method in the first section of the article (information theoretic solution), but I think that reddit is using the entropy of an image instead of the self-information. The self-information yields to better results in our case (empirically). In the latter sections I use different methods.