Stereo correspondence using the Hopfield neural network of a new energy function

Abstract This paper presents an approach using a Hopfield neural network to the stereo correspondence problem for extracting the 3D structure of a scene. The stereo correspondence problem can be defined in terms of finding a disparity map that satisfies three competing constraints: similarity, smoothness and uniqueness. In order to solve the stereo correspondence problem using a Hopfield neural network, these constraints are transformed into the form of an energy function, whose minimum value corresponds to the best solution of the problem, on the Hopfield network. In the process of mapping the constraints into energy function, the energy functions are derived so that the network ensures Hopfield's convergence rule. Stereo correspondence then is carried out through the network evolving energy surface to find the minimum energy corresponding to the solution of the problem. The examples for random-dot stereograms and real images are shown in the experiment, illustrating how the proposed network works.