3D Face Representation And Its Applications

3D human face,as a quite long-standing and attractive topic in computer vision and computer graphic realm,entices many researchers working on it.Based on 3D face data,there are many interesting applications in industry like animation,avatar design in VR etc..However,the stepping stone to achieve these applications is to reconstruct a 3D face.The most popular method when reconstructing a coarse 3D face is recover-ing the parameters of which a 3D face was represented,the coefficients of linear basis specifically.But it cannot meet the requirement when representing 3D caricature that is far deformed away from the normal 3D face.Therefore,we propose an optimiza-tion framework with a brand-new 3D face representation,deformation representation,to tackle image-based 3D caricature modeling problem.With a few annotations on 2D facial landmarks,our algorithm can feedback a satisfying 3D cartoon model.The new representation are constructed on normal 3D face data set while its reconstruction per-formance surpasses that of linear representation on the same data set.Attributes decomposition property is also required in representing a 3D face.Lin-ear representation often decouples a 3D face with identity and expression components,and both components are represented linearly respectively and combined linearly.Al-though it is an easy manner,its identity and expression parts usually entangled.In order to better decompose these two attributes of face,we develop a novel 3D face disentan-gled representation based on deep learning.By treating input data as a graph,the use-ful feature was extracted through graph convolution operators.Our framework renders identity and expression component decoupled better,and benefits several real scenarios,for example,expression transfer on 3D face.