Has anyone used filters in skim to preprocess e.g. the handwritten MNIST digits before classification ? Links: https://en.wikipedia.org/wiki/Optical_character_recognition#Pre-processing lists various methods (with which I'm not at all familiar) http://stats.stackexchange.com/questions/30042/neural-networks-vs-support-vector-machines-are-the-second-definitely-superior/207951#207951
has a picture of some hard-to-classify digits.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)