Dynamic Feedback-based Feature Extraction And Face Recognition Analysis

Face recognition has become a hot issue in many research fields such as machine learning, computer vision and pattern recognition, and it has also been widely applied to the fields of business, law and security for police department etc. Feature extraction is always a basic problem of face recognition which needs to be solved, so how to effectively extract the identification features of face image is the key problem of the face recognition. The key principle of feature extraction is to seed a meaningful low-dimensional representation from high-dimensional data. Although many scholars have proposed many relevant improved algorithm, but there are still many problems and shortcomings. In this paper, some improved algorithms of face discriminate feature extraction are probed aimed at the defects of mainline linear algebra feature extraction methods of the images. Extensive experiments performed on both diverse face databases verify the effectiveness of these proposed methods.The word in this paper including:1、A new feature extraction algorithm based on weighted Fisher faceIt is well known that the traditional PCA algorithm is not consider samples classes information in the process of dimension, and as well as to further maximize the scattering of different classes and concentrate the samples within the same class, three new feature extraction algorithms based on weighted Fisher face are developed:unilateral weighted PCA (WPCA+LDA), unilateral weighted LDA (PCA+WLDA), bilateral weighted method (WPCA+WLDA). The bilateral weighted method which integrates the merits of the other two ways introduce the weight function, which not only consider the relevant information of sample classification in dimension reduction of PCA, but also consider the connection between the various features of all samples and rule out the disturb of correlation between variables. Extensive experiments performed on ORL and AR face databases verify the effectiveness of the algorithm.2、A new feature extraction algorithm based on the dynamic feedback weighted Fisher faceIn order to make use of the useful information in the process of PCA dimension reduction, effectively obtain the samples distribution, and get the best classification in LDA rather than just only separation, a new feature extraction algorithm based on the dynamic feedback weighted Fisher face is developed. This method firstly does PCA dimension reduction to obtain the projection matrix, then the weight is got by continuous feedback information, and weight the covariance matrix to optimize the projection matrix. Finally it adopts weighted LDA further to extract classification features. Extensive experiments performed on ORL and YALE face databases verify the effectiveness of the algorithm.3、A new feature extraction algorithm based on the dynamic feedback of two-dimensional Fisher faceIt is not good of evaluating the covariance matrix, and it is difficult to calculate the best optimal projection space in LDA, a new feature extraction algorithm based on the dynamic feedback of two-dimensional Fisher face is developed. In this paper, the small sample size problem occurred in traditional Fisher discriminant analysis is essentially inevitable. In addition, the developed method is directly based on image matrices. That is to say, it is not necessary to convert the image matrix into high-dimensional image vector like those previous linear discriminant methods so that much computational time would be saved if using the proposed method for feature extraction. This method is mainly to correct the between-class scattering matrix and the within-class scattering matrix of the two-dimensional maximum scatter difference discriminant criterion, and choose the most important supplementary information samples through the dynamic feedback to get the best discriminant criterion. Lastly, Extensive experiments performed on ORL and YALE face databases verify the effectiveness of the algorithm.