Color separation using eigen filtering of a multispectral gray-scale image sequence

The identification and separation of colors or colored objects using only gray-scale images are investigated. A sequence of multispectral gray-scale images is formed by successive imaging through a sequence of optimal filters. Using the signature vector representation of colors obtained from these gray-scale images, the eigen imaging technique is used to generate the eigen filter vector and then the resulting eigen image for separation of one (desired) color from one or two other (undesired) colors. In the eigen filter image produced by this technique, two colors that are not perceptually similar to each other are separable. That is, the gray-scale values of the locations of the desired color are enhanced while those of the undesired colors are suppressed. The signature vectors and the eigen filter vector are not very sensitive to a change in the camera parameter or to the variations of a color.<>