Research On Orthogonal Regularized Kernel CCA Algorithm

With the development of science and technology, high dimensional data frequently appear in scientific research and related fields. In order to make a detailed description of the objective things, people often need to use these high dimensional data. High dimensional data bring people convenience with inches, it also brings the problems: their potential of high dimensional attribute will be found in the law of things behind obstacles, leading to hidden in high-dimensional data behind the rules cannot be directly observed. When people in deal with high dimensional data directly, will meet the so-called "Curse of dimensionality" problem: in the condition of lack of simplifying the data, under the given accuracy accurately function of some variables to estimate,we need the number of samples will be with the dimension of the sample is increased exponentially growth. Therefore, it is very meaningful and necessary to reduce the dimension of the high dimensional data before using it.In this paper, the kernel method is introduced into the regular orthogonal canonical correlation analysis, regular orthogonal kernel canonical correlation analysis method is put forward, which will be regular orthogonal of the scope of application of the canonical correlation analysis algorithm from linear space is extended to nonlinear space. Design and generate simulation experimental data for the canonical correlation analysis of the orthogonal canonical kernel, and compare with the canonical correlation analysis algorithm, which verify the feasibility and effectiveness of the proposed algorithm. Then the algorithm is applied to the feature fusion, and the feature is fused with the method of linear transformation, and the feature of the fused feature is classified by the nearest neighbor classifier. Finally, the algorithm proposed in this paper were applied to the CENPARMI handwritten Arabic numerals and body database and Yale University standard face recognition database experiments are carried out to verify the here and the application of the idea of semi supervised, first of all, the samples are trained to feature extraction and fusion, and then on the test sample classification basis,experimental verification of the algorithm proposed in this paper in feature fusion effect in practical application is superior to the regularized orthogonal canonical correlation analysis.In this paper, the kernel method is first introduced into the canonical correlation analysis of orthogonal regularization, which provides a new method to solve the problem of nonlinear data dimension reduction. At the same time, the orthogonal regularized kernel canonical correlation analysis algorithm is applied to feature fusion and pattern recognition the experimental results of the recognition of the Arabia digital database of CENPARMI and the standard human facerecognition database of Yale University show that it can be applied to practical applications.