RGB to spectral image conversion using spectral pallete and compression by SVD

A basic idea for embedding the spectral pallet in RGB pixels is proposed. A tri-color image is converted into the spectral image, whose pixel is replaced by the spectral reflectance having the closest colorimetric value to that of pixel by looking up the spectral color pallets. SVD representation was very useful for the spectral image compression. We applied SVD for rearranged 36 pixels x36 spectra in local sub-divided block image, so that we could make use of strong correlations in both space and spectral distributions. The spectral image could be very well reproduced from a small number of singular values by SVD. Although a created image has not the real world spectra but pallet-based pseudospectra, the proposed method could be applied to estimate how much the huge spectral image data could be compressed, or to simulate the color appearances for different ink and paper media under the various light sources.