The Study On Key Technology Of Identity And Facial Expression Based On Face Image

Biometric identification has been a research focus in the field of pattern recognition and machine learning. The study of biological feature recognition aims at using pattern recognition and machine learning algorithms, which enable the computer to have a variety of biological characteristics of human beings (such as identity, expression, age, etc.). Face as a human most intuitive biological signal, biometric recognition based on facial images, such as identity recognition (facial recognition), facial expression recognition, age estimation, with very large application prospect, has been the concern of many researchers. This paper mainly focuses on the study of face image based on biometric identification, research focus based on the face image identification (recognition) and the face images of facial expression recognition of the two aspects, respectively for face recognition feature extraction algorithm and cross in the field of color facial expression recognition two challenging problems has done some work, the main research results are as follows:First,an optimal KFKT algorithm based on kernel optimization algorithm is proposed for face feature extraction. The method use discriminant analysis criterion function, using kernel mapping in the sample data sets within class scatter distance to define the objective function of the optimal kernel mapping, and the gradient projection algorithm is used to solve the optimal kernel mapping corresponding to the parameter of kernel function. The optimal kernel mapping parameters obtained by the optimization are applied to the KFKT algorithm, and the optimal KFKT algorithm is obtained. Finally, experiments on FERET and Yale face databases widely used in the world confirm the validity of the proposed method.Second, Three transfer learning algorithms are proposed, which are Kernel Mean Matching, Kullback-Leibler Importance Estimation Procedure and Unconstrained Least-Squares Importance Fitting-uLSIF and Importance-Weighted SVM, which are used to solve the problem of cross domain color facial expression recognition. In this method, the weights of a set of source domain samples are learned by three transfer learning methods, which make the weighted source domain samples have similar characteristics with the target domain. In this case, the IW-SVM will be able to have the ability to identify the target domain samples with the weighted sample of the source domain.Third, Selective Transfer Machine (STM) is proposed, which combines the STM and SVM. Similar to the Second, the aim of this method is to use STM to learn the weight of a set of source domain. But it’s different from the Second work method and that the STM study optimization of the weights is embedded into the SVM, the alternation of STM and the weights of the SVM parameters in the process of learning, learned weights of STM will have better performance.