基于Kinect 2的三维人脸个性化建模

提出一种基于五官形状特征点以及人脸肌肉函数的优化Kinect 2 个性人脸网格的算法,然后通过自适应uv贴图,使结果更符合个性人脸特征.首先,通过提取彩色图的特征点位置获取个性人脸的x,y 空间的特征信息,基本的五官形状特征点为83 个,通过相似三角关系在鼻梁和鼻翼处添加了6 个特征点以丰富五官形状特征信息.肌肉形状特征约束方面,综合肌肉形状约束量有43 个.其次,通过Kinect 2 人脸个性化建模得到中间模型,同时也获得了五官形状和肌肉形状特征的深度先验信息.接着使用基于拉普拉斯算子的微分网格变形算法对Kinect2 的中间模型进行五官优化及肌肉形状约束,通过对比ASM、正则化均值漂移和face + + 算法检测的人脸特征点来选择更精确的五官位置.最后,为了让人脸更具有真实感,还通过平面微分网格形变实现了uv 自动贴图,提升了maya中手动调整uv 坐标的准确性和效率.同时,使用次表面散射光照模型来使人脸皮肤的真实感加强.实验表明,本文所采用的三维人脸模型的个性化变形方法能使Kinect2 模型更好地符合个性人脸特征,这将为后期的表情生成和动画制作奠定良好的基础.

3D Facial Personalized Modeling Based on Kinect 2

This paper puts forward a method based on facial shape feature points and the facial muscle function to optimize the individual face mesh obtained from Kinect 2, then the adaptive UV mapping makes the results more accorded with the character of facial features. First of all, by extracting the feature points of color map to obtain the characteristic information of specific face in X, Y space, basic facial shape feature contains 83 feature points, by means of the similar triangle on the bridge and wings of nose, 6 feature points are added to rich facial features. In the aspect of muscle shape constraint, the volume of muscle shape constraint is 43. Secondly, the modeling of Kinect 2 is used to obtain the Intermediate result, and the depth priori information of the facial features and the shape of the muscles are also obtained. Then, the improved algorithm, namely, the differential mesh deformation based on Laplace operator is used to optimize the facial features and restrict the muscle shape of Kinect2 model. By comparing three algorithms including ASM, regularized landmark mean-shift and face + + to obtain more accurate location of the face feature points. Finally, in order to make the face model more realistic, this paper also realizes the automatic UV mapping by the plane differential mesh deformation, which improves the accuracy and efficiency of manual adjustment of the UV coordinates in maya. At the same time, in this paper, we use the sub surface scattering illumination model to increase the reality sense of facial skin. The experimental results show that the adopted method can meet the individual facial features better satisfactorily than the Kinect2 model, which will provide a good foundation for facial expression generation and animation.