searching for presence of a unique color in security cams to catch a bicycle thief

[atariZen]

If a stolen bicycle has a custom color that is so unique that no other object would likely appear on the surveillance cameras in the city with the same color, could IM be used to find images where the bike appears? Suppose we neglect the fact that the frame is a certain shape, and look simply for images where a few thousand pixels match the very unique color-- could this be done with ImageMagick? Would it be a matter of getting histograms of all surveillance cams and checking whether a significant number of pixels are close to the solid color of the bicycle frame in an image prior to the theft?

[snibgo]

In theory, yes. In practise there are some problems.

Daylight varies in intensity and "temperature", which is roughly the ratio between red and blue wavelengths. The bike might be in shade or sunshine. Bikes are made of tubes, and part of each tube may be in sunshine while another part is in shade. The part in shade will be bluer than the part in sunshine (because it is illuminated by blue sky). And the colour accuracy of cameras will vary.

When searching for the bike, you probably have to ignore the lightness, or at least accept a very wide range of lightnesses. You will also need a fairly wide range for the other two channels (eg ab of Lab, or HC of HCL).