Research On Face Recognition Algorithm Based On Improved2DPCA And Relevance Vector Machine

With the increasing development of modern science and technology,people’s identification has become a trend. Face recognition technology isplaying a more and more important role. Human face recognition technology hasthe advantages of reliability and convenience. It is an important part in the fieldof recognition. On the basis of existing face recognition technology, this paperstudies the main stages of face recognition system further, in order to classify theimage.The main work of this paper focuses on the following aspects: face imagepreprocessing study, feature extraction and classification algorithm study. Firstly,the image normalization and wavelet transform method are used for face imagepreprocessing. The recognition rate of decomposition level of wavelet transformand choice of wavelet basis function are also studied. It improves the defects oftraditional methods, such as large amount of feature number and light sensitiveinfluence. The face images after preprocessing need to be feature selected andextracted. The two-dimensional principal component analysis and the improvedtwo-dimensional principal component analysis (2DPCA) algorithm are studied,and different algorithms in the image dimension and recognition rate werecompared to obtain the best preprocessing and feature extraction of facerecognition method; Secondly, the classification algorithm in face recognition iswidely studied. The basic theory of two multiple classification algorithms,support vector machine and relevance vector machine, are analyzed, and thetraining step and identification method of the two algorithms were described inface recognition, and the "one to one" classification method with high precisionand robustness in vector machine is ultimately selected through the theoreticalanalysis and test.Finally, the simulation experiment on the ORL and Yale face image database is conducted. The influence of training set number in different face libraries onrecognition rate is analyzed, and the effectiveness of applying wavelet transformand improved two-dimensional principal component analysis method in faceimage preprocessing and feature extraction is verified. Through comparisonanalysis in different face libraries, the relevance vector machine method for faceclassification has high recognition rate and better robustness.