A simple idea of RGB to spectral image conversion is proposed. A spectral reflectance with the closest colorimetric value to that of RGB pixel is picked up from the spectral color palette and embedded in each pixel of RGB image. SVD (Singular Value Decomposition) is applied to compress the high-resolution spectral image.Spectral image data are rearranged to 36 pixels × 36 spectra sub-block so that we could make use of strong correlations in both spatial and spectral. The spectral image could be very well reproduced from a small number of singular values by SVD. Although a transformed image has not the real world spectra but palette-based pseudo-spectra, the proposed method could be applied to simulate the color appearances for a given set of ink and paper media under the different illuminants, and to estimate how much the huge spectral image data could be compressed. The paper discusses the color reproducibility by SVD compression and introduces the estimated color appearances for inkjet prints under the different fluorescent lamps.