Research Of 3D Face Modeling Based On The Single Frontal Image

In contemporary society, the 3D reconstruction of human’s face has always been the hot and difficult issue in the image processing and pattern recognition field. In people’s daily life, the 3D face modeling technology has a widely perspect. For example, the large 3D online game and animation rely on the high precision 3D face model, the reconstruction and recognition of 3D face has been the new research area in the field of the video surveillance and the security. In addition, the related field such as Virtual Reality and medical technology are also need 3D face modeling. Apparently, the exploring and research of the 3D face reonstruction exsits both the great scientific value and the practical significance.The traditional 3D face modeling approach has experienced a period which continuous attempts and breakthoughs. Seeking for a higher efficiency and a better reconstruction effect, a promising method named 3DMM was proposed by the scientist. However, the 3DMM need a long-time modeling and exsits some drawbacks like imperfect reconstruction in the ear, so the 3DMM method isn’t used in the realistic scenario. On the basis of the above background information, we studied the 3D face reconstruction which combine the frontial facial image and the morphable model, and also introduced optimization for the unsatisfied result in the ear part. After analyzing the experiment we also discuss the application of the 3D face modeling in the face recognition.Firstly, we deeply analyzing the methods of traditional 3D face modeling technology and the 3D morphable model approach, including their recent status and the technical characteristics. We also showed the advantages and disadvantages of the different methods separately. In addition, we introduced the current sampling situation and the construction level of t the large scale 3D face database in China and overseas. Secondly, aiming at the problem in the traditional morphable model method such as large data processing and complex alignment, we introduced the 3D-TPS and landmarks auto-extraction technology to solve the problems. Meanwhile the possion image blending method was proposed and strived to solve the problem of unsatisfied profile reconstruction result, it boosts the sense of reality in the final result. Finally, after analyzing the performance and the efficiency, we also discuss the 3D face model in the application of face recognition which based on the principle component analysis.The experiment results illustrated that our approach’s result has a better sense of reality especially in the ear part. Meanwhile it has a higher reconstruction efficiency, only need 3.5s average in the MATLAB. The multi-view image which from the reonstruction result improve the rate of face recognition by 6%.