Performance evaluation of 3D hybrid transforms and 2D-set partitioning methods for lossy hyperspectral data compression

Three dimensional nature of hyperspectral data with huge amount of correlation in spatial and spectral domain makes transform coding methods more efficient for compression. Transform methods concentrate signal power in a few coefficients resulting in better low bit rate performance with low computational complexity. A set of 3D hybrid transforms obtained by combining various 1D spectral decorrelator and 2D spatial decorrelator are investigated for their performance evaluation. Wavelet-based methods generate clustered coefficients having parent–child relationship between the subbands. This property can be exploited by entropy encoders to generate bit streams. For entropy encoding, various 2D-set partitioning methods are studied. 2D-set partitioning in hierarchical trees and 2D-tree block encoding exploit parent–child relationship, and 2D-set partitioning in embedded blocks exploits spatial correlation between neighboring pixels within the sub-band in space and frequency of transformed band images. 2D-set partitioning in blocks of hierarchical trees (2D-SPBHT) exploits energy clustering as well as tree structure of wavelet transform simultaneously. It is shown that 2D-SPBHT provides better performance at all the bitrates as compared to other 2D-set partitioning methods irrespective of the 3D transformation used.