Classification consistency for bandwidth compressed multispectral imagery

With the recent surge of interest in machine processing of multispectral imagery for recognition and change detection, the impact of bandwidth compression on the consistency of machine classification is becoming more significant. In this article we evaluate the impact of bandwidth compression on the classification consistency of machine classifiers. The emphasis is on the performance of one-dimensional DPCM and Bayesian classifiers, although we have compared the results with a number of other bandwidth compression techniques, such as the two-dimensional DPCM and the VQ based techniques. Additionally, the effect of the channel error on the classification consistency of the DPCM system is evaluated