Reversible Compression of Hyper-Spectral Imagery through Inter-Band Fuzzy Prediction and Context Coding

A new method for reversible compression of multispectral images is presented. 3-D data are decorrelated by means of an inter-band causal prediction obtained through the fuzzy switching of a set of predictors representative of 3-D micro-patterns occurring across bands, whose coefficients are LS-estimated in a preliminary learning phase. The outcome residuals are entropy coded by exploiting context information. Results of compression on both Landsat TM and AVIRIS images show that the proposed approach outperforms other schemes reported by the most recent and advanced literature.