A remapping technique based on permutations for lossless compression of multispectral images

Multispectral images, such as Thematic Mapper (TM) images, have high spectral correlation among some bands. These bands also have different dynamic ranges. Hence, when linear predictive techniques employed to exploit the spectral and spatial correlation among the bands of a TM image, the variance of the prediction errors becomes greater. Markas and Reif (1993), have used histogram equalization (modification) techniques for lossy compression of multispectral images. In general, histogram equalization techniques are not reversible. However, by defining a monotonically increasing transformation, so that two adjacent gray values will not map to the same gray value of the transformed image, and selecting a target image with a wider probability density function than the source image, one can define a reversible mapping. We introduce a distinct reversible remapping scheme which utilizes sorting permutations. This technique differs from histogram equalization. It is a reversible transformation. We show that, by utilizing the remapping technique introduced and employing linear predictive techniques on a pair of bands, one can achieve better lossless compression than the results reported previously