An object recognition system using self-organising neural networks

Abstract

An object recognition system is proposed using a self-organizing neural network as a basic module for the processing of feature vectors to provide evidence for the recognition state. The modules are integrated to represent various instances of the object scene for which the features are known a priori. The basic architecture of the system proposed was configured to accept a single feature vector or multiple feature vectors at a time. The system was trained on a hypothetical three-object data set for recognition capabilities on object scenes with and without occlusion. The simulation results confirmed the success of the proposed approach.<>