基于Candide模型的人脸深度信息生成技术研究

目前通过2D转3D技术转换的3D电影场景中,绝大部分关于人脸的深度信息生成都不令人满意：深度值不准确、立体效果差。文中为解决特定二维人脸图像转换为三维人脸图像提供了一个比较简单有效的方法。首先,应用Candide模型作为通用的3D人脸模型,然后对通用模型调整控制点进行全局变换和局部变换,通过三维模型与人脸最大限度的匹配得到特定人脸三维网格模型。对网格模型采用顶点规则化和NURBS曲面拟合方法获得光滑的模型表面使其逼近人脸,最后生成准确的人脸深度信息。

Research of Face Depth Generation Information Based on Candide Model

Now in the 2D to 3D conversion of the many movie scenes,the obtained depth information is not accurate and satisfying results in the poor stereoscopic effect.Here a simple and effective approach is provided especially to acquire the fine 3D reconstruction of the specified face in the movie.Firstly the Candide model is adopted as the common 3D face model.Then the global transformation and local transformation is implemented by adjusting the control points of the Candide model in order to obtain the specific face 3D mesh model by the maximal match between the face picture and the model.Finally the specific face mesh model can match better with the 2D face by the mesh＇s vertex regularization and the NURBS surface fitting method,and then the accurate face depth information can be generated from the face mesh model.