Improvements to Palette Feature detector

[dchaplinsky]

At the moment subj is more like a proof a concept.

• It has small palette which I've created from a thin air and modified it until it started to satisfy at least some basic needs.
• It's using PIL (which is not that bad, however if one can come up with openCV solution it'll be also nice).
• It has very simple heuristics (that leads to some false positives) to determine part of the picture, has blue inc features, etc.

The main problem with palette stuff is:

• It seems that palette should be carved with a great care, so we can determine stuff like blue pen marks, yellow/green/pink markers, most popular colours from the MS Office, etc.
• Colour of the background sheet that was used to scan shreds gives all white shreds a little tint (check screenshot below) so it makes it harder to determine real colours and find shreds with similar palette.