Font Size: a A A

Study Of Multi-View Face Recognition Based On Active Appearance Model

Posted on:2012-08-12Degree:MasterType:Thesis
Country:ChinaCandidate:S Y YuFull Text:PDF
GTID:2218330368979469Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
The technology of face recognition is an important branch of biological feature recognition. Because of convenient collection of face images, and face images from different people have great diversity, face recognition technology implements widely, such as in fields of public safety, information security, identity certification etc. Otherwise, pose-variant, illumination and expression affect obviously while face recognizing. It's a very challenging work to recognize face in condition of no restriction. There are many excellent algorithms of frontal face recognition, such as LDA, PCA etc. If one frontal face image can be composed exactly from non-frontal one, we can use these frontal face recognition algorithms to recognize, and then the accuracy of multi-view face recognition can be improved.In this paper, the problems of multi-view face recognition are analyzed and researched comparatively in-depth. Combining the Active Appearance Model (AAM) and Linear Discriminant Analysis, we propose a new multi-view face recognition method, called AAM-LDA. The main summaries of this method are as follows:Firstly, use AAM to modeling multi-view faces from different view of one face respectively, and acquire appearance models of the multi-view faces correspondingly, include shape models and texture models. Secondly, use the models to fit one new input face image, then retrieve the optimal fitting model, and calculate the angle of face image rotation. Thirdly, compose frontal face image through the optimal fitting appearance model. Lastly, LDA face recognition algorithm is used to get the optimal classify feature and recognize the face. Finally, we experiment under Matlab 7.0, and it shows that, the method proposed in this paper works more efficiently than PCA, especially when the angle of face image rotation is large.
Keywords/Search Tags:Multi-view Face Recognition, Active Appearance Model, Principal Component Analysis, Linear Discriminant Analysis
PDF Full Text Request
Related items