A Study Of Face Recognition Algorithm Based On Kernel Learning

With the development of society and economy, the rapid change of Science technology, Face Recognition plays a very important role in many aspects,such as in national defense, security, Internet and life, the human urgently need face recognition system which is stable and efficient to give us more convenient and high quality of life, its main task is to extract effective features of face image, and then it makes full use of the characteristic information to represent face image, to remove redundant information and reduce the dimension of these images, then use the appropriate classifier to realize the human face image classification. This paper will mainly research on feature extraction algorithm of face images based on kernel learning. Mainly from three aspects as follows algorithm research:1 This paper focuses on the matrix singular value decomposition theorem, the stability of singular value, the compressibility of singular value, the scale invariance theorem of singular value, at the same time we set out the most classic "eigenface"-principal component analysis algorithm, and kernel principal component analysis is introduced,in which the input space are mapped to high dimensional feature space,this paper introduces QR decomposition of matrix and its the existence and uniqueness.2 For the KPCA algorithm based on Euclidean distance, the recognition rate is very difficult to get significantly improved and the calculation needs high complexity of the problem, in this paper, the Euclidean distance is extended to the p norm, we first use the approximate low rank decomposition of the Cholesky decomposition to obtain the low dimension kernel matrix, which is quite efficient access key features of face images, then make experiments in the ORL face database and AR, the experimental results fully prove that the new QR-KPCA algorithm based on the p norm has nearly same complexity in the computational complexity, but much better recognition rate in recognition rate.3 With the arrival of the era of big data, processing mass data is particularly important, but if the number of face samples is quite big, dimension of the kernel matrix is staggering, so processing recognition problem that the recognition rate is not high and the calculation problem of large scale is so hard, in order to solve theproblems, This paper presents a fast KPCA algorithm based on hybrid kernel function, which first use mean vector of all the sample vectors to construct the kernel matrix, and then use a hybrid kernel function and a simple computing center of kernel matrix method, so as to obtain the eigenvalues and eigenvector of the covariance matrix, then obtain the effective features of face data. Then we do the experiments on face image in databases ORL, Yale, and finally the experimental results prove that the new algorithm proposed in this chapter achieve better recognition results than other methods, and operation time is shorter, which can be a very good application in face recognition system in real time.