Discriminant Feature Extraction Algorithms Based On Canonical Correlation Analysis

The technology of face recognition is an active subject in the area of pattern recognition. Feature extraction is the elementary problem in the area of pattern recognition. It is the key to solve the problems such as face identification. Canonical Correlation Analysis is a new feature fusion method , which will integration of the two feature vectors into a eigenvector with more identifying information. In this paper,some algorithms of combined feature extraction is improved by mixed Canonical Correlation Analysis into the process of feature extraction,and it proposed a improved canonical correlation.This dissertation focuses on face recognition based on algebraic features with canonical correlation analysis. The main research work and contributions of itare as following:As the traditional canonical correlation analysis can't fully use the category information of the images during the face recognition.In this paper,a new method combines Canonical Correlation Analysis and Maximum Scatter Difference Discriminate Analysis for Feature Extraction is developed.This method not only effectively combines the eigenvectores from the face images but also the information of classes is fully utilized so the correct rate of face recognition is increased much more. Finally, extensive experiments are performed on both ORL face database and Yale face database which verify the effectiveness of the proposed method.Fisher linear discriminant analysis and Maximum Scatter Difference Discriminate Analysis both belong to linear discriminant analysis, and have the same Physical meaning and the similarity process of feature extraction, but as they use the Different discrminant criterion, the eigenvectors reflect the different information about the areas of face images. Fisher linear discriminant analysis and Maximum Scatter Difference Discriminate Analysis are first adopted to extract two sets of features in the same pattern space, respectively. The canonical correlation analysis method is then used to fuse the two sets of features obtained above and to derive more effective canonical discriminant features. Finally, extensive experiments are performed on both ORL face database and Yale face database and experimental results verify the effectiveness of the proposed method.The traditional canonical correlation analysis criterion in defining function of the same group on the correlation between the elements use the multiplication operation, it is a easy problems that it can't guarantee the two groups can be in smallest covariance of the same group of elements at the same time. To solve this problem, it proposed a improved canonical correlation analysis which make a amendment in the denominator of the criteria, it would change the multiplication to the addition, it has a projection correction coefficient by derivation, the projection correction coefficient adjustes the characteristic equations of feature sets and guarantees the two groups can be in smallest covariance of the same group of elements at the same time. The new mothod effectively improves the integration effect and the face recognition rate. In ORL and Yale face databases, the experiments verify its effectiveness.