Research On Face Recognition Algorithm Based On The Discriminant Analysis Of Feature Under Intelligent Environment

With the wide application of face recognition technology in intelligent visual networking,video conferencing systems,public security,financial services and other fields,face recognition technology in intelligent environment has become a hot research topic in the field of pattern recognition.However,in practical applications,the face recognition technology is not only affected by the common factors,but also by the simplification of the feature.Therefore,it is a new research direction to select the face features discriminatively and fuse multiple complementary face features for face recognition in the intelligent environment.The paper studied the face recognition algorithm based on feature discriminant analysis in intelligent environment from three aspects: distinctive feature extraction,distinctive feature fusion strategy and recognition algorithm.The main work is as follows:1.The problem of effective extraction of the face feature under intelligent environment is studied.Aiming at the problem that it is difficult to capture the face recognition information from a single face feature,a improved weighted fusion algorithm of face recognition based on two-dimensional principal component analysis processed by discrete cosine transform and two-dimensional linear discriminant analysis is proposed in this paper,which is according to the complementary idea.This method can effectively overcome the problem that the two-dimensional principal component analysis lacks the discriminant information and the problem that the single face feature is vulnerable to the data noise and the recognition system itself,which can improve the ability of face feature representation.2.The optimal fusion rules of different face features at different fusion levels are studied.In the process of fusing multiple face features for face recognition,the contribution rate of each face feature is not the same,and some face features are more likely to be affected by the number of samples and noise,thus the improved fusion strategy should reduce the weight that they affect the decision results.In consideration of the above situation,an adaptive weight selection method based on the existing fusion methods is proposed in this paper.According to the contribution rate of various face features to the recognition results under different sample numbers,this method can give different weights to face features and improve the accuracy and robustness of face recognition.3.The subspace analysis method based on sparse description in face recognition is studied.Aiming at the problem that the traditional subspace analysis method only uses the information of the training samples to obtain the transform axis and can not represent the test sample well,a face recognition method based on improved traditional subspace analysis is proposed in this paper,which is according to the sparse description idea.And on the basis of the adaptive weight selection method,it is applied to the principal component analysis and linear discriminant analysis.This method can effectively characterize the test samples and improve the classification accuracy.4.Based on the sparse preserving projection,a two-dimensional discriminant supervised locality preserving algorithm is studied.According to the complementarity of two-dimensional discriminant supervised locality preserving algorithm and sparse preserving projection algorithm,a face recognition algorithm based on two-dimensional discriminant supervised locality preserving projection and sparse preserving projection is proposed in this paper,which is on the basis of introducing adaptive balance parameters to construct the objective function that containing data discrimination information and structural information.This method can effectively solve the problem that the two-dimensional discriminant supervised locality preserving projection lacks the global geometric structure information and the sparsity preserving projection lacks the prior knowledge of class labels.