Abstract:

Several statistics of the image graylevel comatrix and graylevel comatrix were first theoretically analyzed. Then a novel Cmean clustering algorithm, which is based on the graylevel comatrix and can be used for synthetic aperture radar (SAR) image classification, was proposed. The characteristics of different statistics were obtained from experiments. The analysis shows that the entropy, variance, the contrast and mean error will perform better. It is shown that when the statistics, which will maximize the betweenclass scatter and minimize the withinclass scatter, is adopted for classification, a much higher performance is achieved.

Key words: graylevel comatrix, synthetic aperture radar, image classification, Cmean clustering algorithm