Face recognition system using principal component analysis and fuzzy artmap

Abstract

Research on face recognition system has been conducted over the past thirty years. The common problem of face recognition systems is catastrophic forgetting where they need to retrain the whole data in order to add a new data. As a result, the training period, processing time, hidden layers and matrix size of input network are increased. This research focused on solving the catastrophic forgetting problem and improving recognition rate. In this thesis, a face recognition system based on Fuzzy Artmap (FAM) as a classifier has been proposed. FAM is an incremental learning approach which offers a unique solution for stability-plasticity dilemma by preserving previously learned knowledge and adapting new patterns. Experiments were conducted to evaluate the performance of both FAM and Multilayer Perceptron Neural Network (MLPNN). The recognition rate obtained were 97.2% and 98.5% using FAM, 90.56% and 81.5% using MLPNN based on local and Olivetti Research Lab (ORL) datasets, respectively. Using FAM, the recognition rate improved by 6.64% and 17% for both datasets, respectively. The results proved that the proposed system offers a solution for catastrophic forgetting and improved recognition rate.