Face Recognition Using Posterior Distance Model Based Radial Basis Function Neural Networks

The success rate of a face recognition system heavily depends on two issues, mainly, i) feature extraction method and ii) choosing/designing of a classifier to classify a new face image based on the extracted features. In this paper, we have addressed both the above issues by proposing a new feature extraction technique and a posterior distance model based radial basis function neural networks (RBFNN). First, the dimension of the face images is reduced by a new direct kernel principal component analysis (DKPCA) method. Then, the resulting face vectors are further reduced by the Fisher's discriminant analysis (FDA) technique to acquire lower dimensional discriminant features. During classification, when the RBFNN is not so confident to classify a test image, we have introduced a statistical method called the posterior distance model (PDM) to resolve the conflict. The PDM is an approach, which takes a decision by integrating the outputs of the RBFNN and a distance measure. We call the new classifier the posterior distance model based radial basis function neural networks (PDM-RBFNN). The proposed method has been evaluated on the AT&T database. The simulation results in terms of recognition rates are found to better than some of the existing related approaches.