|
Just as an important branch of pattern recognition,face recognition have the widespread application demand in the field,such as intelligent monitoring,virtual reality,medicine examination,human-computer interaction and so on.The face recognition has covered computer vision, pattern recognition, artificial intelligence, image processing, the physiology and so on. And the face image varied with its own and the environment condition, Therefore,the research of face recognition is an extreme challenging subject,but must solve many key technologies problems,mainly includes:(1)Face image preprocessing, such as image enhancement, image normalization,elimination of noise and so on;(2) Extraction from face image characteristic,including the partial characteristic and the overall characteristic,may be processed with the statistical method and the geometry method;(3)High performance classification and recognition algorithm, including all kinds of intelligent classification and pattern recognition algorithm and so on.In the reality,limited with some conditions, the engineers is often short of the massive image sample of each people. As far as its dimension is concerned, the number of face sample are very few, therefore the face recognition is a small sample problem. Just as a small sample problem,the traditional classification method,for the one hand,is easy to overfitting study having the poor promoting performance,for the other hand,brings about poor learning performance leading to be unable to process with the very strong classification model of face recogniton.Support vector machines is based on the statistics theory of learning and the minimum structure risk principle, its structure is simple and have strong study performance. The support vector machines method can solve the small sample problem effectively,it may overcome the overfitting studying problem of neural network,on the other hand,its nonlinear ability of classification is very strong. Through introducing kernel function,SVM can change linear inseparable problem into the linear separable one by projecting into the high dimention space. Therefore, support vector machines has become the first choice of classifier for face recognition. Originally,SVM is proposed to solve the two classification problem.Nowadays,the hot research topic is how to apply SVM to the other multi-classification problem,especially,how to construct multi-classification classifier that have very strong learning and promoting performance which is very important for SVM applying to the other fields in the pattern recognition.In the real implementation,the parameter of SVM model is the key factor which affects the classification ability of SVM. Having not theoretical instruction,the traditional methods for parameter selection is the repeated testing one in the experiment. This method needs researcher’s priori knowledge to guide,it also spends lots of time.Therefore,the traditional parameter selection method has limited the development of the SVM theory. For this question, this paper proposes a new method that integrates genetic algorithm and support vector machine,which optimizing the model parameter of SVM by genetic algorithm.Then this method provides a new attempt for the parameter selection of SVM,it can promote the classification effect of SVM,will improve the rate of face recognition by applying this method to the experiment of face recognition.
|