3D Face Model Reconstruction For Face Recognition

This paper presents a method of the automatic three-dimensional facial reconstruction using a single photo feature extraction and the standard model deformation of human face. Of a single frontal face photographs (with some side-angle [-10°, +10°]) for automatic three-dimensional facial reconstruction, using deformable model of a priori statistical knowledge to obtain the depth information, and then most people face model deformation. Just a positive picture, allowing a certain side of the perspective of [-10°, +10°], reducing the requirements for importation of photos to improve the robustness of modeling. Experiments show that this method can generally face rapid automatic photo realistic three-dimensional reconstruction, there are good practical value, can be used for face recognition, animation production and other fields. ASM methods used to improve the precision automatic facial feature extraction point, by making the sparse plane deformation model matching feature points to obtain photographs of human faces depth information, and then the general face model to a specific facial deformation. Based on the color model optimized ASM facial feature extraction, making a certain point of view photos of the side can also have a good effect reconstruction. At the same time, the use of color model based on the texture side of information technology allows integration of the missing issue solved. Experiments show that the method is simple, only a single photo to complete the reconstruction of fully automated, without user interaction to generate three-dimensional model of a better reality. Specific studies on the contents of several parts:1. Based on face detection based on local and global feature extraction.Biological feature extraction is the most recent research topics of concern, which is an important model for the reconstruction of the preliminary work. For local features, this paper, we consider the eye detection, and center ASM algorithm for extracting the initial location, to enhance better do a good basis for convergence. For global features, the paper improved ASM algorithm. In order to improve the texture reconstruction of the effect of the introduction of color model based on the ASM feature point correction, so that feature points do not fall outside the skin, avoiding the side texture reconstruction texture missing phenomenon.2. Put forward a new idea of texture reconstruction. only a positive face image reconstruction of its corresponding three-dimensional model of human faces. For the reconstruction process, the paper will be divided into the shape of reconstruction and texture alluding to two parts. Shape reconstruction of extracting feature points based on the foundation will carry out a good two-dimensional and three-dimensional combination of optimal approximation to obtain three-dimensional feature points, the last under the elastic deformation algorithm to obtain the shape of the model. The texture insinuate, this article based on the color model optimized ASM facial feature extraction to map the pre-correction feature points ASM will be asymmetrical texture information is mapped to the reconstruction of symmetrical model, effective way to avoid non-strictly positive face photos lead to the side of missing texture issues, the reconstruction of a better effect.3. A more humane application framework demo.In order for an early experimental results can be applied to projects and applications, I use the object-oriented C + + language will be the core algorithm and its features are packaged in objects for practical laboratory. At the same time, the MFC in VS 2005 to build the framework of a practical and friendly application platform.4. Corresponds to face model in the application of facial animation.In the reconstruction of a good model, not only on the three-dimensional face recognition system to provide very good way, while in many respects to have a fairly broad application. We use the FDP (Facial DefinitionParameter, facial definition parameters) and FAP (Facial Animation Parameter, facial animation parameters) to achieve facial animation, where the research project the ultimate goal is to generate a good model of human faces into the first to speak, book binding robot robot to help blind reading services for the blind, the paper will study advanced to face the application of facial animation, the future will be applied to the voice on the mouth-type animation.For some in-depth research into the above aspects, this article has made some breakthroughs and progress in several issues, and also access to certain research results.