Local feature-based recognition of partially occluded objects using neural network

Abstract

A new method of recognizing partially occluded objects using neural networks is presented. The neural network consists of a simplified ART-2 and a two-layer feedforward network, and its inputs are the local features of objects. The network is first trained using a set of local features of known objects, then it can be used to recognize unknown object(s). Our numerical experiments using this method show encouraging results, especially for recognizing the occluded objects.