Performance Analysis of Pca Based Techniques for Face Authentication

Accurate authentication is of major concern in real time applications such as an automatic authentication system in any organization. Even though there are many approaches for face recognition in the literature, no algorithm was analyzed with respect to authentication applications. In this paper, we have discussed PCA based approaches Kernel-PCA, Gabor PCA, Phase congruency PCA, Phase congruency-Kernel-PCA, Gabor-Kernel-PCA including classical PCA with Mahalanobis distance measure. The performances of these methods were analyzed with respect to important performance metrics, ROR, EER, and MER. We have also compared the percentage verification rate by varying the percentage FAR. Since we believe that the authentication or the verification rate is highly dependent on the size of available database, i.e., the number of images per subject, we have varied the size of training and testing datasets and accordingly we studied the performance of all the approaches mentioned ahead. All the observed results and graphical analysis of our results were provided in this paper. In our analysis, it was observed that Gabor-Kernel-PCA and GaborPCA approaches shows superior performance in recognition and verification rates with varying size of datasets. Hence these approaches are suitable for accurate authentication applications.