Face Recognition Based On Support Vector Machines

Face recognition in computer is a very active research topic nowadays,shows a wide range of applications in the areas of information security,criminal detection,import and export controls etc. Support Vector Machine is a new machine-learning method and has its unique advantages in pattern recognition because of outstanding learning performance and good capabilities in generalization.The algorithm of training support vector machines with Quantum-behaved Particle Swarms Optimization presented in this paper, namely qpso-svm, is based on the research of training support vector machines with particle swarms Optimization. The algorithm uses the strategy of choosing a working set which decomposes the quadratic programming (QP) problem into a number of sub problems. the working set as the object of the optimization is the Lagranges chosed by meeting the Kohn-Tucker conditions worst. Then optimize sub problems of support vector machines with Quantum-behaved Particle Swarms Optimization ,solve quadratic programming problem. Training the Iris samples with the algorithm of QPSO-SVM, the results of experiments show the possibility of the training support vector machines with quantum-behaved particle swarms, QPSO-SVM is better than PSO-SVM, QPSO-SVM makes up the deficienciy of long running time in training support vector machines with Particle Swarms Optimization.With the advantages of Support Vector Machine in the aspects of treating small samples,high-dimension space and performance in generalization, this paper presents the applications of PSO-SVM and QPSO-SVM in face recognition, providing a new way for face recogniton. Extract the features of face images with the kernel principal component analysis(kpca),then get the eigenvectors of face images as the input of PSO-SVM algorithm or QPSO-SVM algorithm, produce PSO-SVM classifier or QPSO-SVM classifier last. Face images can be recognized with PSO-SVM classifier and QPSO-SVM classifier.The experiments on ORL face database show the feasibility of the algorithms.