On Supporting Identification in a Hand-Based Biometric Framework

Research on hand features has drawn considerable attention to the biometric-based identification field in past decades. In this paper, the technique of the feature generation is carried out by integrating genetic programming and the expectation maximization algorithm with the fitness of the mean square error measure (GP-EM-MSE) in order to improve the overall performance of a hand-based biometric system. The GP program trees of the approach are utilized to find optimal generated feature representations in a nonlinear fashion; derived from EM, the learning task results in the simple k-means problem that reveals better convergence properties. As a subsequent refinement of the identification, GP-EM-MSE exhibits an improved capability which achieves a recognition rate of 96% accuracy by using the generated features, better than the performance obtained by the selected primitive features.