This paper addresses the general image point correspondence problem, with no a priori constraints on camera or object motion. The solution is cast as a cost minimization problem using the simulated annealing algorithm. This is achieved by using a cost function to evaluate the quality of match between pairs of points extracted from each image. Using edge detection and feature extraction techniques points from each image are found which possess strong local features, such as curvature, concavity, standard deviation, and gradient. These features are used to define the cost function for candidate correspondence. The overall algorithm was implemented and tested using a Unix workstation. Test image pairs that were used represented a wide range of relative motions. >
