Pca-based Dynamic Facial Feature Extraction And Its Incremental Learning Algorithm

As a kind of identity authentication technology in the current hot research field , facial recognition technology has had a wide range of application in the information security , access control , intelligent control and other fields . Based on the classic high-dimensional small samples as the research object, facial recognition technology also has stringent requirements in the performance of pattern recognition algorithms . For facial recognition system , the feature extraction of face image is one of the most important algorithms . Especially extracting the effective and stable identification information of face image , not only can reduce the system complexity , but also can improve the recognition rate. At present , because of its simple algorithm to extract face image characteristics effectively , the Principal Component Analysis (PCA) has been more widely applied in face recognition field . However, there are also many technical bottlenecks in the feature extraction algorithm of the Principal Component Analysis (PCA) , such as the prevalence of low accuracy and rate, high algorithm complexity and bad real-time , poor training and so on , which are especially the prominent problems in incremental learning . In the current facial recognition , incremental learning is based on the sample vectors of the operations , which would lead to a loss of recognition efficiency and real-time , especially during the process of nuclear operations . In view of the above questions , the article has done the in-depth study , and the definite research contents are as follows :(1)Aiming at the incremental learning problems in the the feature extraction of dynamic facial recognition, this paper puts forward a dynamic facial feature extraction algorithm, which is based on the two-dimensional incremental learning algorithm, and the whole operation process is based on two-dimensional image matrix calculations . First of all , through the original samples , we construct a projection space for the two-dimensional principal component analysis , into which any samples can project . When the new samples join , and by the parameters of the original projection space and new samples' information , we can create a new project space . For that , this can complete the incremental learning problems based on two-dimensional vector . All the experiments prove that , the incremental learning algorithm of principal component analysis based on two-dimensional image matrix , not only can ensure a good recognition rate , but also can reduce the computational complexity and improved the system recognition efficiency to some degree .(2)when nuclear operations are done in dynamic facial recognition system , for the high computational complexity of image vector operation , the article proposes a facial feature extraction algorithm based on incremental kernel principal component analysis . It is the incremental algorithm nucleation of the two-dimensional principal component analysis . Firstly , through two-dimensional kernel principal component analysis , it takes the feature extraction of sample image . During the process , all the parameters of nuclear matrix is based on the image matrix calculations , so it can do the incremental learning for new samples . This learning algorithm can solve the problems of high computational complexity , then effectively reduce the training time by not changing the leaning accuracy .(3) For the incremental learning algorithm based on two-dimensional principal component analysis and Two-dimensional kernel principal component analysis, this article respectively applies them to facial recognition system for simulation experiment. In addition , every experiment is separately on the basis of JAN-A dynamic face database and ORL standard face library . At last , all the experimental results show that : This system can keep the high recognition rate , at the same time , it can also reduce the sample training time and improve the system recognition performance to some extent .