Prediction model selection for compression of satellite images

One major problem of lossless image compression is the lower compression ratio obtained. This is due to the wide spatial bandwidth of image pixel intensities. In this paper we describe an approach for reduction of satellite image spatial bandwidth thus improving the compression ratio. In this approach we code image pixels in a predetermined sequence, predicting each pixel's intensity using a fixed linear combination of a fixed constellation of nearby pixels, then coding the prediction error. Computer experiments have been performed on various satellite images to evaluate the performance of different prediction models on improving the compression ratio.