NURBS Warping For Facial Expression

The three dimensional facial expression generation is one of the hottest research topics in computer vision.Through the joint efforts of many scientists for decades,3D facial expression generation algorithm has vigorously developed,and there are a large number of real world applications and potential commercial applications: facial animation,3D face-assisted recognition and so on.However,in the actual research process,there are still many challenges and improvements in 3D facial expression generation algorithm,such as how to deal with the 2D images from the different perspectives,illumination and pose,the local fine description of facial expressions and so on.These aspects put forward higher requirements for the robustness and accuracy of 3D facial expression generation algorithm.In order to describe the geometric structure of 3D facial expression,this paper proposes a 3D facial expression generation algorithm based on NURBS and face feature points.The 3D face reference model represented by the NURBS surface is generated from the 2D face images set.Through adjusting the face feature points,the reference model owns the expression;During the adjusting process,we use the geometry constraint with second derivative that keep the stability of the change of facial shape,and smooth constraint with Gaussian weight that ensure that the part around the feature points is smooth;Meantime,we use the NURBS line to curve the mouth contours,which brings the reference model more realistic expression.To validate the proposed method,the experiments are performed on the standard image,the image from LFW data set and video sequence,respectively.The results well indicate that the effectiveness of the proposed method for generating a 3D facial expression shape from the standard image,the unconstrained image and video sequence,the proposed algorithm has the robustness and generalization.In summary,this proposed algorithm precisely curves the generated 3D face expression shape and improves the robustness and accuracy.In the future research work,we will apply the generated results into the 3D assisted face expression recognition,facial expression transfer and so on,this paper lays a good theoretical foundation for the next work.