A Research Of Feature Extraction And Application Of Face Recognition

In recent years, the face recognition technology is widely used in computer vision,artificial intelligence, which makes the technology have important application value and practical significance in real life. Such as, entrance guard system, public security,etc. At the same time facial recognition technology becomes the most direct, friendly way of contact in the field of biometric identification, which makes it become the active topic in the research of computer pattern recognition field. The most classic face recognition algorithms, whose theory has reached a very mature state, is presented by the Person K and Fisher R A principal, who name them as component analysis(PCA)and linear discrimination analysis(LDA). And, so far the two methods has become the benchmark of many emerging method.This paper, on the basis of the earliest and the more mature of theory facial feature extraction algorithm PCA and LDA, mainly does the following work:This paper, first introduces the development history of facial recognition technology and its application field, shares research status at home and abroad, at the same time makes the brief elaboration about face database used in face recognition technology, and analyzes the main technical difficulties affecting the efficiency of face recognition.Then, the concrete process of PCA and LDA algorithm and how to apply to face recognition system are presented. The two methods to example analysis of high-dimensional data dimension reduction is given through the experiment. The specific process of PCA + LDA algorithm are given by combination of the advantages of the two methods respectively.These methods all belong to the linear feature extraction method, can’t extract the high-dimensional nonlinear structure information of the face image.The third and fourth chapter are the core content of this article, the third chapter analyzes the KPCA algorithm, which is proposed by combining PCA with kernel function coming true the nonlinear changes in support vector machine, makes up for the lack of the linear feature extraction method. The improving methods based on KPCA is used to extract face nonlinear structure information. At the same time, this paper shows the manifold learning algorithm LLE and the improved algorithmprinciple of LDA. The extensive experiments are carrying out on ORL face database,the analysis of the advantages and disadvantages of each method are given.However, PCA method can only extract the global information of samples,not consider the identification information between the samples, combining PCA with Fisher discrimination analysis method, and considering the global information and identification information, sample recognition rate are promoted, but it cannot form effective map of nonlinear structural information of face image.and in terms of the nature of several manifold algorithms,they are all kind of linear feature extraction method.As for the nonlinear characteristics of face higher dimensional distribution, this paper proposes a new face recognition method based on kernel principal component analysis and minimum distance differential projection. This method first extract facial geometric features by kernel principal component analysis, obtain the corresponding projection matrix to reduce samples dimension, subspace includes the non-linear information of sample after dimension reduction, the minimum distance identification was implemented in subspace, projection matrix not only contains the identification information between samples but also the neighborhood relationship between samples.Finally, relevant experiments are implemented at the university of Cambridge ORL, the U.S. military FERET face database and YALE face database to verify the effectiveness of the method.