The request for efficient compression of hyperspectral images becomes pressing. A new lossless compression algorithm based on 3D contexts prediction for hyperspectral images is presented. Spectral band grouping algorithm is introduced to divide hyperspectral images into groups according to the neighboring band correlations

then band reordering is performed for each group. The important bands containing large information can be determined by using adaptive band selection algorithm

on which clustering is carried out according to the spectral vectors. 3D contexts are defined based on the neighboring causal pixels in current band and the corresponding colocated causal pixels in reference band. Combined with the clustering results

the optimal predictive coefficients of each cluster are trained respectively. Experimental results show that the proposed algorithm can give better lossless coding performance.