A membrane-inspired bat algorithm to recognize faces in unconstrained scenarios

Face recognition under unconstrained environments has become increasingly important due to the broad prospect in real-world applications. In order to counter uncertainties imposed by faces captured in such unconstrained imaging situations, a robust, discriminative and computationally efficient feature selection scheme is of paramount significance. In this regard, bio-inspired feature selection methods have been exploited due to their sophisticated ability, flexibility and adaptability. However, their performances tend to deteriorate severely in large-scale domains such as face recognition due to the premature convergence problem. In this paper, high-dimensional LBP features are extracted from face images and fused with Gabor wavelet features using Canonical Correlation Analysis (CCA). To further enhance the discrimination power of the facial representation and to alleviate the curse of dimensionality, a novel membrane-inspired feature selection approach is proposed, where a Binary Bat Algorithm (BBA) under the framework of Membrane Computing (MC) is employed. Inherent parallelism and non-determinism are two distinguishing characteristics of MC that can help in maintaining the diversity of population and balancing the exploration exploitation trade-off. In the proposed membrane-inspired BBA (MIBBA), the structure as well as the evolution, dissolution and communication rules of MC are integrated into the BBA to enhance the trajectories of bats. Furthermore, the Great Deluge Algorithm (GDA), is integrated into the skin membrane to further improve its exploitation ability. Experimental results show that the proposed approach yields competitive recognition rates and outperforms well-known state-of-the-art methods on three benchmark databases (AR, LFW and GBU). Further experimental evaluations justify the ability of the proposed approach to handle the small sample size problem. (C) 2017 Elsevier Ltd. All rights reserved.