基于正交图像生成人脸模型的合成分析方法

针对采用现有基于图像的人脸建模方法生成的三维人脸模型存在的人工性缺陷问题,提出了一种基于合成分析方法的个性化三维人脸建模方法.利用合成方法由两幅正交人脸图像生成一个初始三维人脸模型,比较基于颜色直方图方法合成的人脸图像纹理与输入图像纹理的差异,根据纹理差异来指导对人脸网格的局部自适应细分,不断调整合成的三维人脸模型,从而更好地保持了人脸的精细细节特征.实验结果表明,使用该细分反馈算法,可以减少模型的人工性缺陷,在整体上提高了合成人脸模型纹理的真实感.

Analysis-by-synthesis approach for facial modeling based on orthogonal images

Some residual ghosting or blurring artifacts may be visible in 3D facial model synthesized by the image-based facial modeling algorithms. This work proposed an analysis-by-synthesis approach for image-based personalized 3D facial modeling. Starting from two orthogonal face images of an individual, synthesis technique was employed to obtain its initial 3D facial model. A color histogram based technique was then used to measure the mis-registration error between the synthesized texture and the original orthogonal images' texture. Through using this error to guide the local and adaptive subdivision step, the 3D facial mesh model was refined to better preserve fine facial features. Experimental results show that the artifacts are reduced after the feedback and local adaptive subdivision steps; and that the output texture is more photo-realistic.