New Technique for DCT-PCA Based Face Recognition

Abstract

Automatic face recognition is a high level computer vision research. As faces are complex natural stimuli that differ dramatically, hence developing a computational approach for accurate face recognition is very difficult indeed. In this paper a robust face recognition system is developed and tested. Here 2D Discrete Cosine Transform (DCT) is exploited for feature extraction and certain normalization techniques are invoked that increase its robustness to variation in facial geometry and illumination. The DCT coefficients of face images are truncated in Gaussian or exponential way. Then without doing inverse DCT Principal Component Analysis (PCA) is applied directly for dimensionality reduction. Lastly face recognition task is performed by K-nearest distance measurement. Only a few non-zero eigenvalues related eigenvectors are taken for this method and experimental tests on the ORL face database are performed which give outperform result with respect to existing techniques. This method achieved cent percent result when the prob image is applied from the same database.