Discriminant Feature Extraction Algorithms For Face Images Based On "Small Sample Size" Problem

The technology of face recognition is an active subject in the area of pattern recognition. There are broad applications in the fields of law, business, security for police department etc. For the particularity and complexity of the face image, face recognition is also the very difficult problem. There is still much work to do before applying the technology in our daily lives. Feature extraction is the elementary problem in the area of pattern recognition. It is the key to solve the problems such as face identification. In this paper, some improved algorithms of face discriminate feature extraction are probed aimed at the defects of mainline algebraic feature exaction methods of the images. Extensive experiments performed on both diverse face databases verify the effectiveness of these proposed methods.The work in this paper including:It is well-known that the applicability of Fisher linear discriminant analysis(LDA) to high-dimensional image recognition tasks such as face recognition inevitably suffers from the so-called"small sample size"(SSS)problem. In this paper, We propose a novel discriminant analysis method to essentially avoid the SSS problem using scatter difference discriminant criterion. Difference from LDA, The method adopts the difference of between-class scatter and within-class scatter as discriminant criterion rather than the ratio of them. Extensive experiments performed on both ORL and AR face database verify the effectiveness of the proposed method.A new method of discriminant analysis based on scatter difference criterion in residual space is developed in this paper. It successfully avoid the instability of face images because of different illumination. What is more, it utterly solves the so-called"small sample size"(SSS) problem usually occurred in traditional Fisher linear discrminant analysis by using a new maximum scatter difference discriminant criterion. Finally, extensive experiments performed on both ORL face database and yale face database verify the effectiveness of the proposed method.Cosidering the so-called"small sample size"(SSS) problem in nature and the"inferior"problem in traditional Fisher linear discriminant analysis, a new method of feature extraction based on modified maximum scatter-difference criterion is developed in this paper. The method gives an effective way to resolve two difficulties of the traditional Fisher linear discriminant analysis theoretically in face recognition. Finally, extensive experiments performed on ORL and AR face database verify the effectiveness of the proposed method.In this paper, a novel image projection discriminant analysis based on scatter difference criterion is developed for image feature extraction. In the proposed one, 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. In this paper, by analyzing the equivalence between two-dimensional maximum scatter-difference discriminant analysis and maximum scatter-difference discriminant analysis begin with the intrinsic essence of image projection analysis, a new method is improved. It constructs the image matrix over again in a new mode, so that the extracting discriminant feature would be more advantaged for classification. Finally, extensive experiments performed on ORL and AR face database verify the effectiveness of the proposed method.In this paper, a novel two-dimensional kernel maximum scatter difference discriminant analysis (2D-KMSDA) is developed for the extraction of nolinear feature by using the well-known"kernel trick". It extracts much more effective nolinear feature and made the true recognition rate improved saliently. What's more, it also offers a unified framework for two-dimensional nolinear discriminant analysis. Finally, extensive experiments performed on AR face database verify the effectiveness of the proposed method.According to traditional canonical correlation analysis (CCA), a novel method of combined feature extraction called two dimensional canonical correlation analysis (2DCCA) is proposed in this paper. It combines feature matrix directly by using the main idea of image projection in face recognition. Compared with traditional CCA based on feature vectors, this method has the following two main advantages: first, the small sample size problem (SSS) occurred in traditional CCA is essentially inevitable as a result of the evidently reducing dimension of the covariance matrix. By the same reason, the second advantage is that much computational time would be saved if using the proposed method. Finally, extensive experiments performed on ORL and AR face database verify the effectiveness of the proposed method.