Nonlinear Predictive Vector Quantization of Multispectral Imagery

Multispectral satellite images of the earth consist of sets of images obtained by sensing electromagnetic radiation in different spectral bands for each geographical region. A new compression method for such data sets is proposed where a small subset of image bands is initially vector quantized. The remaining bands for the same spatial region are estimated from the quantized images by a nonlinear predictor which is optimal for the mean squared error distortion measure. The residual (error) images are conditionally encoded at a second stage based on the magnitude of the errors. This scheme exploits both spatial correlation (by vector quantization) and spectral correlation (by nonlinear prediction) inherent in multispectral images. Simulation results on an image set from the Thematic Mapper with 7 spectral bands are presented. A substantial improvement is obtained by using the nonlinear predictor over the optimal affine predictor. Image compression ratios between 20 to 30 are achieved with remarkably good image quality.