Face Recognition Based On Nonlinear Statistical Reduction Dimension And Mirror Image Features A Combination Of Parity

As a unique, the most common and easy for people to accept biometrics, human face has complicated structure and very expressive, and the face photographic process will be affected by many factors just as surrounding light, photographic angle, photographic distance and so on, so the face recognition algorithms performance would be depended on how effectively extract the natural characteristics of face images; and because of face images are high-dimensional data, it would easily encounter "curse of dimensionality" problems. So the face recognition algorithms are focus on how to fully and effectively extract natural facial features, how to effectively reduce the dimensionality of the data and circumvent "curse of dimensionality" problems. This article is based on the latest domestic research results, by studying and improving the face recognition dimensionality reduction algorithm, so that the algorithm would fully and effectively extract natural facial features, and the recognition efficiency of the algorithm will be improved too. The main work and achievements include the following aspects:1. Proposed an improved face recognition method—A Face Recognition Method Based On Combinational Mirror-like Odd And Even Images Features. Using the mirror image transform and principle of the parity image decomposition extract the odd and even mirror-like symmetrical images, and then extract the eigenvectors of odd and even mirror-like symmetrical images respectively, and combined those eigenvectors into a new combination eigenvector reasonably, the new combination eigenvector will extract features of the face images fully and effectively, thereby improving the recognition performance of the algorithm.2. Further study the face recognition method based on combinational mirror-like odd and even features with noise adding in the images, and analysis how rations of odd mirror-like symmetry image eigenvectors and even mirror-like symmetry image eigenvectors in the new combination eigenvectors in this case.3. Because of the limitations performance of face recognition method based on KPCA with a single kernel function, proposed an improved face recognition method—A Face Recognition Method Based On KPCA with combination kernel functions. Combined those single kernel functions whose have complementary characteristics into a combination kernel function, the combination kernel function will learn characteristics from each single kernel function and its integrity could be better than all of those single kernel function, and the robustness and generalization of the algorithm will be greatly improved, too. This article will combine Gaussian kernel functions with linear kernel function and polynomial kernel function into new combination kernel functions respectively, and apply those combination kernel functions to face recognition method based on KPCA. The experimental results show that the face recognition method based on KPCA with combination kernel functions outperform greatly than PCA and single-kernel-based KPCA methods.