Generalized sloped facet models useful in multispectral image analysis

In order to identify targets, it is interest to examine the signature in two or more wave bands and to correlate the energy intensities in these bands. One needs to develop algorithms for systematically analyzing the components of a target or an image, in general, in order to classify it properly. Sloped facet modeling constitutes a first-order approach to such description. Combined with multi-spectral image processing, a powerful technique is potentially available for image understanding. The quad-tree concept, together with split-and-merge techniques, can be used in combination with sloped facet designs in order to accelerate the identification process. The purpose of this paper is to unify these ideas. In addition we present a new method based on the Sobolev norm which makes careful, use of gradient information in order to derive linear forms having better variance properties.