Near lossless compression is a trade off between lossy compression and lossless compression.Usually

K-L transformation is used for spectral decorrelation

and DC transformation is used for spatial decorrelation

when multispectral images were near lossless compressed.In this paper

we analyzed the correlation of multispectral images

classified the spectral correlation as statistical correlation and structural correlation.Then we proposed a newtechnique: K-L transformation was used for statistical spectral decorrelation

and prediction tree technique was used for structural spectral decorrelation and spatial decorrelation.This technique removed spectral correlation effectively and gave a better experimental result than K-L-T and DCT.