PARAMETER OPTIMIZATION OF SUPPORT VECTOR MACHINE USING ENHANCED HYBRID PARTICLE SWARM OPTIMIZATION IN NON-LINEAR FACE AUTHINTICATION PROBLEM
One of the well-known machine learning methods is support vector machine (SVM). With small number of training samples, it can discover the global optimal solutions for the complex non-linear problems such as face recognition. However, choosing the optimal parameters of SVM is a big challenge which has a high impact in the classification results. “Particle Swarm Optimization (PSO)” has been used to find the optimal parameters of SVM but PSO has drawbacks in inertia weight selection which is fixed number and in population initialization which is random. In this paper, a new face recognition technique based on Hybrid Particle Swarm Optimization and Support Vector Machine (LOPSO-SVM) is introduced. The hybrid PSO algorithm based on Logarithm decreasing inertia weight and opposition particle swarm initialization which can improve the convergence speed in PSO. Principle Component Analysis (PCA) has been used for feature extraction process and the extracted features was passed to the proposed technique. In the experimental results, human face database CASIA V5 is utilized to verify the performance of face recognition technique LOPSO-SVM. The proposed technique is compared with PSO-SVM and AOPSO-SVM. The experimental results shows that the proposed method gave higher face recognition accuracy than PSO-SVM and AOPSO-SVM and outperform the other method in finding the optimal parameters of SVM.