Abstract: Aiming at the problem that the huge amount of data of remote sensing images causes great pressure and low compression ratio to transmission and storage, an efficient lossless compression method with improved adaptive band rearrangement and minimum mean squared error prediction is proposed. This method can adaptively determine the optimal order of the bands, and can make full use of this sort correlation to eliminate image redundancy based on the minimum mean squared error prediction. Firstly, the method adaptively groups the hyperspectral image bands and uses the minimum spanning tree algorithm to sort within each group to improve the inter-spectral correlation of adjacent bands. The intra-band bands are then adaptively selected for inter-spectral and intra-spectral predictions, removing redundancy from hyperspectral images. Finally, the binary arithmetic encoding of the prediction residuals removes statistical redundancy and completes the lossless compression of hyperspectral images. Experimental results based on ZY1-02D hyperspectral images show that this method effectively utilizes the correlation of intra-spectral and inter-spectral, improves the prediction performance, superior to common compression methods.

Key words: hyperspectral images, lossless compression, correlation, minimum spanning tree, predictive coding, entropy coding