Particle swarm optimization based block feature selection in face recognition system

Face is one of the most widely used and accepted biometric traits. Face recognition systems are now being utilized in many applications ranging from individual (e.g., smartphone user authentication) to large scale (e.g., border crossing screening) scenarios. Most face recognition systems employ feature selection after feature extraction to enhance the accuracy of the frameworks. In other words, feature selection is one of the important phases that any recognition system must go through as the final results depend on it. Thus, in this paper, we present an optimized feature selection method based on Particle Swarm Optimization (PSO) to select a block of feature instead of single feature to ensure the distinctiveness and variations of features with application to face recognition system. In particular, first the captured face image is divided into a regular number of blocks (sub-images), then Binarized Statistical local features (BSIF) local descriptor is applied on each block for feature extraction. Next, a PSO scheme is utilized to select the blocks/features. The nearest neighbour classifier is employed to get the value of the fitness function (here, equal error rate (EER)) for block/feature selection. The blocks with the smallest EERs are chosen to represent the face image representation and recognition. Experimental results on public ORL faces database show promising results. The proposed face recognition system obtained EER equals to 1.028% with only 4 blocks out of 16, and recognition rate up to 93.5%. While the system was able to obtain an EER equals to 0.5% and recognition rate = 97% using 8 blocks out of 64 blocks.

Automated summary

This paper introduces an optimized face recognition system with a feature selection method based on Particle Swarm Optimization, which enhances accuracy by selecting blocks instead of individual features. Experimental results show promising performance on a public face database.