An Optimization Approach to Edge Reinforcement

The present investigation is concerned with an optimization approach to edge reinforcement which does not involve probabilistic concepts. A cost function is defined, based on the local pattern of edge responses, which is low for sharp, strong edges, and a steepest descent method is used to adjust the edge magnitudes so as to reduce the value of this cost function. The edge reinforcement results obtained in this way appear to be better than those obtained by relaxation methods, whether based on conditional probability or on optimization. The results of an application of the considered procedure to a portion of a Landsat scene are shown, taking into account the input image, the results of 10 iterations of the steepest descent procedure, results produced by means of the relaxation process reported by Schachter et al. (1977), and results obtained with the aid of the optimization relaxation process considered by Faugeras and Berthod (1979).