ILLUMINATION-INVARIANT RECOGNITION OF TEXTURE IN COLOR IMAGES

We represent texture in a color image by using spatial correlation fun ctions defined within and between sensor bands. This representation ha s been shown to be useful for surface recognition, but the structure o f spatial correlation functions depends on the spectral properties of the scene illumination. Using a linear model for surface spectral. ref lectance with the same number of parameters as the number of classes o f photoreceptors, we show that illumination changes correspond to line ar transformations of a surface correlation matrix. From this relation ship we derive a distance function for comparing sets of spatial corre lation functions that can be used for illumination-invariant recogniti on. This distance function can be computed efficiently from estimated correlation functions. We demonstrate, using a large body of experimen ts, that this distance function can be used for accurate surface recog nition in the presence of large changes in illumination spectral distr ibution.