3D Face Modeling Based On SVM

In recent years, computer graphics technology had been rapidly development.3D facemodeling technology as one part of the fields of computer graphics, has been received widelyconcerns of many researchers.3D face modeling has a very wide range of applications inmany fields such as film, advertising, games, communication, medicine, intelligentrecognition and so on. These all have inextricably relations with3D face modelingtechnology.Currently, there are two main existing3D face modeling approaches. The first method of3D face modeling was based on three-dimensional scanning device. But it is huge in size andis very expensive. Furthermore, it’s very complex in processing data. Therefore, it has beenlimited of popularization and application. The second kind is based on images. This kind ofmethod does not require expensive equipment, and just can complete the task only with theordinary camera. So this method is the research focus now.The method of3D face modeling based on SVM has been studied in this paper. Thismethod is one of the3D face modeling based on images. The main contents include threepoints. The first, it needs to get the points of the three-dimensional features for the humanface. The second, we should make the model of the3D face by the SVM fitting. Finally, thetexture mapping is necessary. This paper has been proposed a new method which in the aspectof accurately getting the3D feature points of the face. And the method was based on thebinocular stereo vision and the orthogonal images combined. We obtain the3D feature pointsusing the method of orthogonal images when they are both visible in the positive image andthe side image because its process is simple relatively. Otherwise we use the method based onbinocular stereo vision. After accurately obtained the3D feature points, we use them trainingthe SVM learning machine and then fit the feature points of unknown. Followed, we can drawthe3D face geometry model with all feature points in three-dimensional space. Finally, a facetexture mapping method is proposed based on positive image and skin color model combined.In this method we first using a frontal face image to project to the establishment3D facegeometry model vertically, and then correct the wrong color area in side face using the facialskin color model. And a realistic3D face model is produced.