A modified discriminant sparse representation method for face recognition

Recently, a new discriminative sparse representation method for robust face recognition that uses ℓ
₂
-norm regularization was reported. In this paper, direct data-driven calculation of the balance parameter used in the objective function is presented. The modified system preserves the advantages of the original method while improving the recognition accuracy and making the system more automated, i.e., less dependent on the user's input. Extensive simulations are performed on six face databases, namely, ORL, YALE, FERET, FEI, Cropped AR, and Georgia Tech. Sample results are given demonstrating the properties of the modified system.