Class-based compression algorithm for hyperspectral images

The huge amount of hyperspectral images creates challenges for data storage and transmission, thus it is necessary to employ efficient algorithm for hyperspectral images compression. An efficient lossy compression algorithm based on spectral classification was presented in this paper. The C-means algorithm was performed on the hyperspectral images to realize the unsupervised classification. According to the classification map, an adaptive Karhunen-Love transform was performed on each class vector with the same spatial location in the spectral orientation to remove the spectral correlation, and then two dimensional wavelet transform was performed on each principle component. In order to achieve the best rate-distortion performance, the embedded block coding with optimized truncation coding was performed on all the principle components to produce the final bit-stream. Experimental results show that the proposed algorithm outperforms other state-of-the-art algorithms.