Clustering Regression Wavelet Analysis for Lossless Compression of Hyperspectral Imagery

Abstract

Recently, Regression Wavelet Analysis (RWA) was proposed as a method for lossless compression of hyperspectral images. In RWA, a linear regression is performed after a spectral wavelet transform to generate predictors which estimate the detail coefficients from approximation coefficients at each scale of the spectral wavelet transform. In this work, we propose Clustering Regression Wavelet Analysis (RWA-C), an extension of the original 'Restricted' RWA model which may be used to improve compression performance while maintaining component scalability. We demonstrate that clustering may be used to group pixels with similar spectral profiles. These clusters may then be more efficiently processed to improve RWA prediction performance while only requiring a modest increase side-information and computational complexity.