COMPUTATION REDUCTION OF THE MAXIMUM-LIKELIHOOD CLASSIFIER USING THE WINOGRAD IDENTITY

The maximum likelihood classifier is one of the most used image proces sing routines in remote sensing. However, most implementations have ex hibited the so-called ''Hughes phenomenon'' and the computation cost i ncreases quickly as the dimensionality of the feature set increases. B ased on the above reasons, the recursive maximum likelihood classifica tion strategy is more suitable for hyperspectral imaging data than the conventional nonrecursive approach. In this paper we derive some comp utation aspects of quadratic forms by applying the Winograd's method t o three previous approaches. The new, modified approaches are approxim ately four times faster than the conventional nonrecursive approach an d two times faster than the existing recursive algorithms. Copyright ( C) 1996 Pattern Recognition Society.