Feature point matching of affine model images using hopfield network

Abstract

This paper presents an approach to match feature point of a pair of 3-dimensional affine model images. The affine transferring parameters are computed by a set of corresponding feature points, which are obtained based on 2D Hopfield neural network. The design of energy function of the neural network optimizes the matching error of the feature points. Two affine geometric constraints, epipolar and homography are used without the restriction to scene's particularity. A pair of affine model images tests the performance of the proposed method. © Springer-Verlag Berlin Heidelberg 2005.