EBFS-Fisher: An efficient algorithm for LDA-based face recognition

This paper presents a new algorithm for LDA-based face recognition with selection of optimal principal components using E-coli Bacterial Foraging Strategy (EBFS). A GA-PCA algorithm has been reported to find optimal eigenvalues and corresponding eigenvectors in LDA. In their paper, a fitness function has been proposed to find the optimal eigenvectors to be used in LDA using a Genetic Algorithm (GA). However, the crossover method used results in differences in offspring and mutation never allow us for a physical dispersal of child in a chosen area. This may not help us in finding optimal eigenvectors for improvising accuracy of face recognition algorithm. On the other hand, this paper proposes a new algorithm called EBFS-Fisher which uses a nutrient concentration function (cost function). Here the cost function is maximized through hill climbing via a type of biased random walk. The proposed EBFS-Fisher algorithm offers two additional advantages. First, the proposed algorithm can supplement the features of GA. Second, the random bias incorporated in EBFS help us to move in the direction of increasingly favorable environment. The Yale data base is used for evaluation. Experimental results depict the fact that about 3% (Rank 1) improvement can be achieved as compared to GA-Fisher.