THE ILLUMINATION-INVARIANT MATCHING OF DETERMINISTIC LOCAL-STRUCTURE IN COLOR IMAGES

The availability of multiple spectral measurements at each pixel in an image provides important additional information for recognition. Spec tral information is of particular importance for applications where sp atial information is limited. Such applications include the recognitio n of small objects or the recognition of small features on partially o ccluded objects. We introduce a feature matrix representation for dete rministic local structure in color images. Although feature matrices a re useful for recognition, this representation depends on the spectral properties of the scene illumination. Using a linear model for surfac e spectral reflectance with the same number of parameters as the numbe r of color bands, we show that changes in the spectral content of the illumination correspond to linear transformations of the feature matri ces, and that image plane rotations correspond to circular shifts of t he matrices. From these relationships, we derive an algorithm for the recognition of local surface structure which is invariant to these sce ne transformations. We demonstrate the algorithm with a series of expe riments on images of real objects.