Face recognition using spectrum-based feature extraction

This paper proposes a spectrum-based approach for enhancing the performance of a Face Recognition (FR) system, employing the unique combination of Discrete Fourier Transform (DFT), Discrete Cosine Transform (DCT) and Binary Particle Swarm Optimization (BPSO). Individual stages of the FR system are examined and an attempt is made to improve each stage. DFT and DCT are used for efficient feature extraction and BPSO-based feature selection algorithm is used to search the feature space for the optimal feature subset. Experimental results, obtained by applying the proposed algorithm on Cambridge ORL, Extended Yale B and Color FERET face databases, show that the proposed system outperforms other FR systems. A significant increase in the recognition rate and a substantial reduction in the number of features is observed. Dimensionality reduction obtained is around 96% for ORL and more than 99% for Extended Yale B and Color FERET databases.

Figure optionsDownload as PowerPoint slideHighlights
► DFT+DCT based feature extraction for FR irrespective of dataset.
► Significantly reduced number of features using DCT+BPSO.
► Highly improved recognition rate even with a simple Euclidean classifier.