3D Face Reconstruction Based On A Single Image

3D face reconstruction has a strong representative and broad application prospects, it has becomes hot topics of research fields such as computer graphics, computer vision and pattern recognition in recent years. Meanwhile, 3D face reconstruction is also cross-cutting issues included medicine, psychology and other fields. Research and solve on this subject can provide theoretical guidance for complex modeling, and plays a important role in the development of many more subjects.Article first work done on the previous summary of the following areas: 1) Introduced the previous reconstruction methods according to chronological, and then classification introduced the important image databases and graphics databases. 2) Discussed the method to establish the 3D face dense point correspondence: Compared the method of using 2D Thin Plate Splines function on 2D cylindrical coordinate system and the method of using 3D Thin Plate Splines function on 3D cartesian coordinate system to establish the dense point correspondence. 3) compared to the reconstruction algorithm to obtain the appropriate algorithm for this article: introduce the method of the way to minimize the error rate, and calculate the weights of the sample face graphics, to obtain the target face graphics. Then compared the method of the way to using 2D thin plat function on the 3d face model to deformation again. At last , obtained the latter method is more suitable for the reconstruction of this article.In this paper, complete the following tasks: 1) Detailed analysis of the Candide face model, marked each vertex in the X-Y coordinate system of candide face model, to facilitate future work; 2) East-Asian face images used in this article, calculated the vertical and horizontal ratio relationships between this image and candide model; 3) Mark the feature points on the 2d face image, and establish the corresponding relationship between the feature points on 2d image and the key vertex of 3d graphic, and deformated the key points of 3d model, obtained a rough match between the image and the model; 4) On the basis of the cooresponding relationship between the feature points on 2d image and the key vertex of 3d graphic, combinated the key vertex of 3d face model with 2d thin plate spline function to calculate the coordinates of the other non-feature points, and obtained the target face graphic. 5) Written a application with VS2008 and Direct3D interfaces, obtained the test result through the program. The result shows that the proposed method can create a more realistic 3d human face. Finally, use this model to compare the previous reconstruction results by other researchers.