Draft: Using YCbCr rather than LUV background detect as it is PIL native

Very experimental, for big picture comments.

Started with LAB rather than YCbCr as it uses the same L channel as LUV. However, the PIL implementation of LAB appears unstable, with the A channel sometimes wrapping through the UINT8 providing numbers that are way off (such as one 0 in a plain background image when most of the A channel is 240ish).