Palette reordering under an exponential power distribution model of prediction residuals

Palette reordering is one of the most effective approaches for improving the compression of color-indexed images. Recently, a theoretically motivated modification of a reordering technique proposed by Zeng et al. (2000) was suggested (Pinho et al. (2004)), based on an exponential distribution model of the prediction residuals. In this paper, we develop this theoretical analysis further, exploiting a broader model based on exponential power distributions.