Research On Face Recognition Based On Affine Hull Multi-manifold Distance Metric

Face image conveys a lot of important identity information.In the pattern recognition of computer,face image recognition is a challenging research topic.In this paper,the local weight matrix and the domain adaptation method are combined with the multi-manifold distance metric method which based on affine hull respectively.The two methods are applied to face recognition.The experiment on the standard face dataset shows that the two methods in this paper can obtain better effect of face image recognition.The main works of the thesis are listed below:(1)A new method of local neighborhood based on multi-manifold metric learning for face recognition is proposed.For small sample size problem in face recognition,the face image is pre-processed by using the feature face.The Euclidean distance is used to select the nearest neighbors of each data point in the pre-processed face data set,The local weight matrix is obtained by this and the error distance between the reconstructed data points and the original data points is calculated.At the same time,using the image set modeling manifold,the affine hull is used to represent the dataset information of manifolds and calculate the distance metric matrix among manifolds.By maximizing the distance between the manifolds and minimizing the distance between the data points and the reconstructed data points to find the dimension matrix.(2)A novel approach of face recognition via domain adaptation and manifold distance metric learning is proposed.This approach aims to construct a projection space in which major information of original data can be preserved as much as possible to reduce influence of disturbing elements in data.The maximum mean discrepancymethod is used to reduce the difference between training set and test set,and the affine hull is used to measure the distance between manifolds.And next,we find a projection space for the constructed metric model.The maximum mean discrepancy is minimum and the manifold distance is maximum in this projection space.(3)After the original data is projected onto the projection space,the KNN classifier is used to determine the category of the unknown face image.On the one hand,the research of this paper can improve the recognition rate of face images.On the other hand,it also explores the application of domain adaptation and manifold learning to face recognition.