Ultraspectral image compression using two-stage prediction: Prediction gain and rate-distortion analysis

Abstract

The sheer size of Atmospheric Infrared Sounder images, a type of ultraspectral cube that includes over two thousand spectral bands, is such that their compression is of critical importance. A traditional approach to this goal is by combining reversible preprocessing, where image redundancy is better exposed, with a pure prediction stage that performs compression at a cost of introducing some controlled distortion. In this paper we focus on the effect of using a prediction stage that integrates both, linear prediction (LP) and a search procedure, as a way to obtain better quality. Since it can be seen that this additional search stage does not affect the compression rate, its only drawback is from the computational point of view, making algorithm optimization a key factor. In addition, we introduce a mechanism to dynamically select the LP filter order such that when combined with two-stage prediction the overall rate distortion is greatly improved.