Lossless compression of multispectral images with interband prediction error deltas

This is a study in the use of spectral deltas for lossless multispectral image compression. When bands are correlated, the spectral deltas are typically smaller than prediction errors. Therefore, we want to devise methods that use spectral deltas when there are regions of correlation and use prediction errors when there are not. We apply our methods to a variety of image sets and use one of the best compressing still image algorithms as a benchmark.