Face Recognition System with Automatic Training Samples Selection Using Self-organizing Map

The paper deals with evaluation of automatic training samples selection method based on self-organizing map (SOM) in face recognition systems. In earlier paper [1] we presented an approach for automatic training samples selection using various clustering algorithms with good results on the CMU PIE face database. We showed that with the use of SOM we can achieve a good training samples selection. In this paper we further evaluate this approach with the use of face recognition systems based on principal component analysis (PCA) and support vector machines (SVM). We compare the results with random (uncontrolled and controlled) training samples selection and we evaluate the recognition accuracy of each method.