Hand Contour Classification Using Differential Evolution Algorithm with Ensemble of Parameters and Mutation and Crossover

Abstract

Biometrical identification of persons using the contour of a human hand belongs to a very interesting and not yet totally explored areas, and its accuracy and effectiveness depends, to some extent, on the technical possibilities in scanning of persons. The presented paper solves the problem with use of a combination of various methods. A hand contour, topological description of the hand, evolutionary algorithm, and linear regression algorithm to estimate correct knuckles positions is used. For comparison of geometrical data, the Iterative Closest Point (ICP) algorithm is used in its genuine shape. Just the modern evolutionary optimizers enabling to change from ground the view how to solve similar problems but at the expense of higher algorithm development demands. However, it enables to cut down computational demands of ICP algorithm markedly. Experimental verification of proposed methods was performed with use of two different databases THID and GPDS with persons of different gender and age (c. 20-65 years) with total number of persons in individual databases 104 and 94. Experimental results proved very successfully the suitability of use the combination of the methods ICP and evolutionary optimizer called EPSDE for solving the given task with final algorithmic complexity O(N) and successful rate at classification given by coefficients THID:EER=0.38% and GPDS:EER=0.35% on real images.