Enhanced Face Recognition Using Discrete Cosine Transform

In signal processing, important information is mainly required and processed. Discrete Cosine Transform (DCT) provides a great compaction capabilities. One of the challenges of face recognition using DCT and any other algorithm is poor illumination of the acquired images. In this paper, anisotropic diffusion illumination normalization technique (AS) and DCT were used for recognition. The AS was employed as a preprocessor before applying the DCT as a feature extractor. The new face recognition technique named 'ASDCT' was assessed on ORL, Yale, and extended Yale-B databases. Performance metrics were generated and evaluated by verification and identification rates using nearest neighbor classifier (NNC). Appearance based techniques were also exploited for comparison with the new method, and results show that 'ASDCT' outperformed many renowned algorithms in the literature. It has produced up to 93.4% verification rate at False Acceptance Rate (FAR) =0.1% on ORL database. Therefore, its performance using both controlled and uncontrolled databases is considerably good. It is believed that, the new framework enhances the DCT feature extraction for more efficient face recognition.