基于特征点加细的多分辨率人脸形变模型及人脸建模

Multi-resolution morphable model and face modeling based on features refinement

提出基于特征点加细的原型三维人脸对应方法建立多分辨率人脸形变模型,并根据该形变模型的特点使用多分辨率模型匹配方法由单张正面人脸图像进行三维人脸建模。该方法以人脸模型上的眼、眉、口、鼻等主要几何特征为基准点标注基础网格,然后通过加细特征点网格完成原型人脸之间的对应,进而建立多分辨率的形变模型;根据形变模型的构造特点,把待匹配图像按照与模型相同方式进行加细,然后进行多分辨率的人脸模型匹配。实验结果表明,新的对应算法可以有效地实现原型三维人脸之间的对应,能够克服传统的光流对应算法对应效果差、算法精度低的缺陷,提高形变模型的精度。新的匹配算法不仅能够提高模型的匹配速度,而且可提高模型匹配的效率和精度,缩短模型匹配的时间。

An algorithm of aligning prototypical 3D faces based features refinement was proposed to establish multi-resolution face morphable model, which was used to rebuild 3D face modeling through multi-resolution model matching method, This algorithm took eye, eyebrow, mouth, nose, etc. as main geometric features to mark base grid, and then it aligned 3D faces and established face morphable model by using refinement. According to the characteristics of the morphable model, the authors refined an input image in the same way as the model, and then rebuilt face model through multi-resolution matching. The experimental results show that the new alignment algorithm can align the face prototypes well and improve the precision of the morphable model; the new matching algorithm can accelerate the speed of model matching, improve the precision and efficiency of model matching, and shorten the time of model matching.