Face recognition system invariant to expression

Abstract

This paper proposes an efficient face based recognition system which is also invariant to expression. Face is a popular trait for automatic recognition but variation in facial expression tends to produce large distortion and changes in certain regions of the face and this may reduce the accuracy of any good recognition system. It has been observed that most of the facial distortions tend to be concentrated around the mouth region of the face. The system ignores all areas of ambiguity and a highly effective Self-Organizing Map (SOM) based network is used to extract the true features present in the remaining regions of the face. k-NN ensemble method is used to classify faces to find the accurate matching of the subject. The system has been tested on the IITK database and the FERET database and Correct Recognition Rate (CRR) is found to be more than 91%. © 2014 Springer International Publishing Switzerland.