Semi-Paired Multi-set Correlation Projection Algorithm And Application

Face recognition is a research hotspot in the field of pattern recognition,and it is alos an important research content in biometric recognition.Face feature extraction is a key part of face recognition,which has been widely studied at home and abroad.In practical application,the same person's face image can be characterized by different view features.Due to factors such as missing data and asynchronous sampling,multi view data of face images can not do one-to-one correspondence among different views in real application scenarios.The existing literature refers to this kind of data as semi-paired multi-view data.At present,compared with traditional multi-view feature extraction,there are few researches on feature extraction algorithms for semi-paired multi-view data.Based on this,this paper focuses on semi-paired multi-view data,combines the neighborhood relations among different views,and researches on feature extraction algorithms based on neighborhood association and face recognition system.The major research contents and achievements of this paper are as follows:(1)A multi-set neighborhood correlation analysis algorithm(MNeCA)is proposed.The algorithm makes use of a large number of unpaired samples by using neighborhood information between different views,which can alleviate the"overfitting" problem caused by few paired samples effectively.In order to improve the performance,two regularized MNeCA algorithms:PRMNeCA and LRMNeCA are developed.PRMNeCA and LRMNeCA can utilize unpaired samples in both multi-view and single-view ways.Several experiments on face datasets:FERET,Yale and AT&T show that the proposed algorithms can extract discriminative low-dimensional features.(2)A deep multi-set neighborhood correlation analysis algorithm(DMNeCA)is proposed.This algorithm is a nonlinear extension of MNeCA,which can obtain nonlinear features through the transformation of neural network,and can effectively deal with the complex nonlinear relationship between semi-paired multi-view data.The effectiveness of the proposed algorithm is verified on MNIST dataset and F-MNIST dataset.(3)By using the proposed feature extraction algorithms,a face recognition system based on neighborhood association representation is designed and implemented.The system mainly includes face image acquisition module,face image detection module,face image preprocessing module,neighborhood related feature extraction module and face recognition module.Through the real environment test,it is proved that the system can effectively classify the face images.