Feature Selection Using Dynamic Binary Particle Swarm Optimization For Enhanced Iris Recognition

Iris Recognition (IR) is a demanding field, owing to varying contrast and live-tissues. The important contrast invariant features need to be extracted to address this problem. This paper proposes a novel feature selection evolutionary algorithm, namely, Dynamic Binary Particle Swarm Optimization (DBPSO) for enhanced IR. DBPSO generates a highly optimized global best vector, using which, the number of selected features not only gets reduced but also the most important feature subset gets selected. Images from two benchmark iris databases, namely IITD and MMU, are preprocessed and subjected to feature extraction using Gabor Filter and Discrete Cosine Transform using different geometric shapes, following which feature selection using the proposed DBPSO is performed to obtain feature vector space for optimal feature subset. Experimentation done on the databases shows an overall significant increase in the average recognition rate and a substantial reduction in the number of features selected.