Multispectral imagery is one of sequence images. Usually

the frame compression of sequence images is based on the decorrelation between frames. To multispectral imagery

there is less correlation between frames. So the conventional frame compression is invalid. In this paper

we present a new lossless compression technique that based on classification to decorrelate spectral correlation and spatial correlation

we called it classification～based spectral prediction and spatial prediction (CBS2P2) method. The imagery was classified first

then was compressed by spectral prediction and spatial prediction. We created spectral pridictors with spectral vectors

for the gray values of image pixels were based on spectral vectors

and chosed the second mode of JPEG standard rediction modes for lossless compression as spatial predictors. The experiments with 6 bands SPOT data

5 bands NOAA-AVHRR data and 6 bands TM data show the compression ratios can be improved.