Performance Analysis of the Feedforward and SOM Neural Networks in the Face Recognition Problem

This paper presents a comparative study between a feedforward neural network and a SOM network. The paper also proposes the incorporation of a new spatial feature, face feature lines, FFL, to represent the faces. FFL are considered as new features based on previous studies related to face recognition tasks on newborns. Besides the face feature lines, the feature vector incorporates eigenvectors of the face image obtained with the Karhunen-Loeve transformation. A face recognition system is based on a feedforward neural network, FFBP, method. The second classification scheme uses a self organized map, SOM, architecture combined with the k-means clustering algorithm. Experiments comparing both architectures show no significant differences for the ORL database, 92% for the FFBP and 90% for the SOM. However results obtained for the Yale database, 60% for the FFBP network and 70% for the SOM, indicate a better performance with the SOM architecture