Surface Orientation Invariant Matching of Spectral Signatures Kurzfassung

For monitoring purposes, change detection and segmentation of remotely sensed multispectral images it is necessary to normalize the images in order to yield spectral signatures independent of illumination, atmospheric conditions and surface orientations. For small scale artiicial objects in high resolution images however, the surface orientation is not known, and hence comparison of spectral signatures is prone to error due to varying illumination. We assess the relative contribution of direct versus diiuse illumination, and present a spectral distance measure and a spectral transformation which provide color constancy and are independent of the degree of direct versus diiuse illumination and thus surface orientation.