Lossless Predictive Compression of Hyperspectral Images

Abstract

We have presented a number of different predictive coding schemes for the compression of hyperspectral images. While the schemes differ in the details of their implementation their outline is essentially the same. Each algorithm tries to approximate the ideal case of stationary data for which optimal predictors can be computed. The approximations can be viewed as attempting to partition the data space into sets within which the stationarity assumption can be applied with some level of plausibility. The extent to which these algorithms function or fail depends upon the validity of their assumption. There is clearly much more work to be done before we can claim that the problem of predictive lossless compression has been solved.