Abstract: A neural network based approach to feature point correspondence in a long image sequence is proposed in this paper. The proposed approach formulates the correspondence problem as a constrained optimization problem and uses a Hopfield neural network to find the solution. The problem is viewed as that of optimizing a suitable energy function which is chosen by taking into account attributes including Euclidean distance between two feature points in two consecutive frames, prediction of feature points using filter, occlusion of points, and smoothness of point motion. This approach is robust especially for occlusion. At the same time, this method provides the two-dimensional motion parameters, so it may be used in two-dimensional motion analysis and predictive visual tracking. Experimental results on a synthetic sequence and a real image sequence of human motion show that the correspondence can be established efficiently, although it has been previously proved to be a combinatorially explosive problem.