A Research Of Human Face Detection Based On AGPSO-AdaBoost Algorithm

As an emerging discipline, face recognition have received increasing attention, so face detection which is the position link of face recognition is also received more and more attention. Today, the face detection is not only used to solve the problem of computer vision, but also used in commercial and law enforcement sectors, such as suspect photos matching, user authentication, user login control, group monitoring and intelligent human-computer interaction.Currently, most classic face detection algorithm is face detection based on AdaBoost algorithm proposed by Viola and Jones in2001. The algorithm brought new development and progress to the research of face detection problem, but there are some shortcomings, such as the relatively higher false detection rate and the relatively lower recognition rate of the weak classifiers. In this paper, we present an AGPSO-based AdaBoost Algorithm, which is based on AdaBoost algorithm and Supporting Vector Machine. Because face detection based on AdaBoost algorithm should put the weak classifiers to the linear combination into strong classifier, so this is a combinatorial optimization problem. The current position of the particle is weak classifier weights, if these weights are described by the standard particle swarm optimization algorithm is, it is difficult to define the particle’s speed. Therefore according to the basic idea of particle swarm optimization combined with genetic algorithm, proceed weak mutation and weak crossover operation, make the current solution cross with individual optimum and global optimum, the resulting solution would be used as the new location of the particles in the solution space.The algorithm mainly aims at tentatively dealing with the problems of the traditional AdaBoost face detector’s poor in the bad Generalization ability, the optimization of the weak classifiers’weight coefficients, the lower detection rate and the higher false alarm rate. The introduction of mutation operator, crossover operator and simulated annealing operations, increase the diversity of particles, in order to make algorithm achieves the balance of the global search ability and local search capabilities, resulting in a global optimal solution which is the global optimal weight coefficients of weak classifiers. In the AGPSO-based AdaBoost Algorithm, two primary features are concluded as follows:(1)Construct AdaBoost weak classifiers by using SVM.(2)The algorithm, which combines the PSO with genetic algorithm and utilizes crossover operator and genetic operator and adopts simulated annealing algorithm as adaptive strategy, globally optimizes the weight coefficients of weak classifiers.Experimental results demonstrate that the detection rate and the ability of generalization of human face detection are all improved by utilizing the AGPSO-based AdaBoost Algorithm in comparison to traditional AdaBoost Algorithm and PSO-AdaBoost Algorithm, meanwhile, it has low false alarm rate.