Spectral-spatial hyperspectral image compression based on measures of central tendency

Hyperspectral images have become an active research topic due to their higher spectral resolution provided by dense spectral sampling at each pixel by a number of narrow and contiguous bands of wavelength. In this paper, we propose a lossy compression approach that uses a novel technique of applying central measures to exploit inherent spectral correlation in consecutive bands of hyperspectral images and use of vector quantization on transform coefficients to exploit spatial correlation in order to achieve higher compression. It is generally perceived that use of compressed hyperspectral images may affect the results of post-processing stages such as classification and unmixing, however this possible adverse effect has been considered in this algorithm by the use of a spectral distortion measure, Spectral Angle Mapper (SAM) along with conventional Peak Signal to Noise Ratio and Compression Ratio to evaluate performance of the algorithm.