Project Analysis For Face Recognition Based On Ensemble Learning

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. Since the generalization ability of an ensemble could be significantly better than that of a single learner, ensemble learning has been a hot topic in international community of machine learning. In this paper, the mainstream feature extraction methods and ensembled approach are researched and improved. And this paper made useful explorations on a combination of feature extraction and ensemble in face recognition. Extensive experiments performed on both diverse face databases verify the effectiveness of these proposed methods.The work in this paper including:This paper proposed enhanced Fisher discriminant criterion based on artificial interference according to generation of Fisher discriminant criterion. This discriminnant criterion is based on user-defined similarity measure of samples. This similarity measure of samples is attached to the within-class scatter matrix and the between-class scatter matrix generated Fisher criterion. It maximizes within-class similarity and minimizes between-class similarity when Fisher criterion minimizes the distance of within-class and maximizes the distance of between-class. Then a better scatter matrix is got. Extensive experiments performed on both ORL, AR and AR_Gray face database verify the effectiveness of the proposed method.In order to make the diversity between the classifier, as well as the effective use of such diversity, this paper presents a classifier ensemble method Cagging based on class information. Training sets of each classifier generated through selecting samples repeatedly based on class information enhance the diverse of each classifier. And, combining the classify result of all classifiers by voting with weight vector to use the diverse of each classifier better. This weight vector is set for each classifier according to its classification performance to each class. Finally, extensive experiments performed on both ORL face database verify the effectiveness of the proposed method.The classifiers ensemble is an important approach designing high performance pattern classification system. This paper proposed EPCA algorithmic for getting the effective single classifier. The algorithmic gets lots of PCA projecting transformation are gotten from RSM, so lots of original classifiers are gotten. According to their classification performance in each pattern class, their preservation scores are given. Total score decided by all classes orders the preferential rank for classifiers preservation, by which a set of classifiers is selected from original classifiers. Finally, extensive experiments performed on ORL verify the effectiveness of the proposed method.For linear discriminant analysis (LDA), different training data imply different within-class covariance matrix and between-class covariance matrix, and different optimal discriminant vectors. In this paper, associated LDA with ensemble learning, a novel linear discriminant analysis approach named ELDA is proposed, which utilizes boosting to help generate a set of optimal discriminant vectors. The optimal discriminant vectors generated from Foley-Sammon linear discriminant analysis (FSLDA) hold maximal pattern separability in the context of Fisher criterion and less redundancy information by reason of their orthogonality in each other. Finally, extensive experiments performed on ORL and AR face database verify the effectiveness of the proposed method.