Abstract: According to hyper-spectral images having strong correlation both in spectral and spatial, a novel compression scheme based on second order difference predictive that combines with spatial predictive was presented. MED predictorwas used to remove the spatial correlation. Second order difference predictor was used to remove spectral correlation.Then a unified predictor was designed based on the weight of predictive error. At last near lossless compression was completed after context based coding. The results show that the compression ratio can reach up w 12. 7 when the PSNR is about 39413, so the algorithm is efficient.

Key words: Hyper-spectral images,Decorrelation,Lossless compression,Predictive coding