Face Recognition Using Discrete Cosine Transform and Nearest Neighbor Discriminant Analysis

TLDR: A new combination of DCT with Nearest Neighbor Discriminant Analysis (NNDA) for face recognition is proposed, found to be robust for expressions and small pose variations of facial images.

Abstract: In this paper we have proposed a new combination of DCT with Nearest Neighbor Discriminant Analysis (NNDA) for face recognition. Discrete Cosine Transform (DCT) is a powerful transform to extract features from a face image. It is requisite to discriminate classes using extracted DCT features. Some low frequency DCT coefficients are selected and given as input for Discrimination analysis. We used DCT for feature extraction, low frequency DCT coefficients are selected since they carry most of the information, then NNDA is used for discrimination analysis. We applied 2-level Discrete Wavelet Transformation(DWT) only for non-match faces and smoothed those images by zeroing vertical coefficients of DWT, since those coefficients are responsible for the effect of small expressions and edges in facial images, considering this, image is reconstructed after zeroing its vertical DWT coefficients and classified once again. When experimented, we achieved 99% (at 50 features) and 98.5% (at 70 features) recognition rate on ORL and Yale databases respectively. This method is found to be robust for expressions and small pose variations of facial images.