Object representation based on contour features and recognition by a Hopfield-Amari network

In this paper, a type of Hopfield-Amari neural network is built based on a so-called curve bend function (CBF) for recognition of planar shapes (contours). Two kinds of features, the real-valued features and binary features, are defined by means of the CBF for given contours to characterize the shapes. The overlap between features are reduced effectively in the process of a network construction The experimental results demonstrate that the proposed system is powerful and reliable in solving shape recognition problems.