Lossless compression of hyperspectral image based on spatial-spectral hybrid prediction

This paper proposes an improved lossless compression algorithm based on the spatial-spectral hybrid prediction. We choose the prediction modes between the spatial and the spectral domains by computing the local correlation coefficient. If such coefficient is larger than the pre-designed threshold, the spectral linear predictor is adopted, which is able to capture more spectral correlation by re-estimating the correlation. Otherwise, MED predictor is used. Finally, prediction error images are coded by RICE algorithm. Experiments are carried out on AVIRIS scenes. Simulation results show that the proposed method outperforms 3D-CALIC algorithm, MED and GAP spatial lossless prediction algorithms.