Research On Synthesis Methods Of The Frontal Face Image

The frontal face image synthesis is to recovery the information to synthesize thefrontal face image technology based on one or more profile face images. The frontalface synthesis is a process to build a data from scratch. It can use one or multiple profileface images to restore the information of the cover part of the face, and to obtain anaccurate frontal face image.In this paper, the profile face deflection angle estimation and the frontal facesynthesis algorithm are researched. Firstly, the principal component analysis methodand the establishment and fitting process of the active appearance model are elaboratedand analyzed in detail, and the multi-angle active appearance model is presented. Next,a method of the profile face deflection angle estimation is introduced. Then, linearobject class theory is elaborated in detail. On the basis of the theory, the relativelyimportant information of the face and the visible texture information of the profile faceare used to improve the frontal face image synthesis algorithm. The frontal facesynthetic results are improved.The accurate estimation of the profile face deflection angle is the key problem forthe frontal face image synthesis. In order to estimate the angle accurately, the activeappearance model is used to modeling multi-view faces from different view of one facerespectively, and the appearance models of the multi-view faces are acquired, whichinclude shape models and texture models. The models are used to fit one new input faceimage, and the optimal fitting model is retrieved, and the face image deflection angle iscalculated accurately.The linear object class theory is the theoretical fundamental of the frontal facesynthesis algorithm. In order to synthesize the frontal face image effectively, first of all,the shape and texture information of the face are separated to get the shape and texturereference information. Then, the shape and texture reference information are used topredict the shape and texture information of the frontal face. At the same time, thevisible texture information of the profile face is also used to synthesize the frontal face.The linear object class theory and the principal component analysis method are usedfully in the process of the frontal face synthesis. The face can be regarded as the linearobject class, so it has the properties of the linear object class. Thus, the face can berepresented linearly by a group of face in the same perspective, and the corresponding linear weight is the same. Because the dimension of the face information is very large,the principal component analysis method is used to reduce the dimension of the faceinformation to remove the unimportant information and retained the key information.The proposed frontal face synthesis algorithm has carried on the simulationexperiments. The experimental results show that the proposed algorithm is feasible andthe synthetic frontal face image is satisfactory.