Study For AdaBoost Algorithm Based On PSO And Its Application In Human Face Detection

Face recognition is one of the most popular biometric technologies, which is widely used in authentication information technology field. During the past 10 years, as one of the branches of biometric technologies, face recognition, which has the characteristic of portability and noninvasive, has aroused worldwide concern in the research area of pattern recognition and artificial intelligence along with the rapid development of network communication and multimedia technologies as well as the fast improvement of computer hardware level.One of the critical steps of face recognition is face detection. While face detection is based on face feature extraction in order to check whether there exists face from a given picture (usually grayscale image).If there exits face then mark the location and size of face individually. Among the various face detection algorithm the most famous one is proposed by Viola which is known as AdaBoost Face Detection. However, there are some aspects of AdaBoost which need to be improved such as the upgrading of detection rate and the reduction of false alarm rate and the training time of weak classifier.For the problems above, this paper takes AdaBoost and Supporting Vector Machine as the base, and tentatively suggests an improved AdaBoost Algorithm based on PSO, which tries to deal with the problems of the traditional AdaBoost's beyond the optimization of the weak classifier coefficients. In the proposed algorithm, three essential technologies are adopted as follows:1) construct AdaBoost face detector by using SVM. 2) Optimize the weak classifier coefficients with PSO.3) Improve PSO premature convergence defects by using simulated annealing.Experiments are conducted with OpenCV and Matlab as experimental tools and Yale face database, some pictures after pre-processing are taken as the experimental material to test the PSO-AdaBoost Algorithm. The Result reveals that the PSO-AdaBoost Detector has the ability to performance better than the traditional AdaBoost Detector in the detection performance indicators.