Face Feature Selection and Recognition Using Separability Criterion and Binary Particle Swarm Optimization Algorithm

Discrete cosine transform (DCT) is an effective method to extract proper features for face recognition. Discrete cosine transform can only map the resource data to another data field instead of compress data. How to select the DCT coefficients that are most effective for classification is an important problem. This paper proposes a novel method to search the best discriminant combination of DCT coefficients. A feature selection algorithm according to the separability criterion is used to preselect the DCT coefficients, and then follows a search algorithm based on binary particle swarm optimization and support vector machine to find an optimal combination of the DCT coefficient. The performance of the algorithm is assessed by computing the recognition rate and the number of selected features on ORL database and Cropped Yale database.