Research On 3D Face Pose And Expression Correction

In the field of machine vision,three-dimensional modeling technology has been widely studied by both domestic and abroad scholars.3D face reconstruction is one of the key subfields of this field,and has a lot of matured applications in gaming,medical,filmdom,television and commerce.Due to the fact that changes in the facial postures and expressions cause deformations,the recognition rate of 2D face recognition systems decreases,therefore,it is necessary to perform corrections on these postures and expressions.In this thesis,a method of face alignment(feature point location)based on Active Appearance Model(AAM)in 2D face images,and a facial pose and expression correction algorithm based on 3D Morphable Model(3DMM)are proposed.Firstly,face alignment is the basis of face reconstruction.In this thesis,the local texture model and the AAM model are combined in the 2D face image to obtain 68 facial feature points.Then,the Fast-SIC algorithm is adopted to improve the AAM model and to enhence the fitting efficiency.The algorithm achieves good alignment effects both on AFLW and LFPW,which are the two famous large face databases,as well as self-build face database.Then,based on the face alignment results,3D face reconstruction is performed.A BFM-3DMM model,combine expression parameters into classical 3DMM model,is proposed.However,the face corrected by the BFM-3DMM model is not smooth enough.Due to the fact that the SFS algorithm is not constrained by the original statistical model,this algorithm is applied to re-correct 2D face from BFM-3DMM model.The experimental evaluation results show that the corrected 2D faces have smoother apperance and higher fidelity compared with classical 3DMM model,and can also retain image background information.Finally,considering the drawbacks of the 3DMM model for Asian face features,a 3D reconstruction system for asian faces by employing a surface structured light 3D scanner is analyzed.Processes including calibration,scanning and smoothing steps are presented in detail.Smoother 3D face modeling results are obtained,which lay the foundation for an improved 3D facial model suitable for asian faces.