| 3D face recognition is one of the hottest research topics in the field of computer vision,image processing and pattern recognition.Compared with the traditional 2D face recognition,3D face recognition has many advantages,which could solve the problem of poor robustness caused by the changes of posture and illumination.However,it is difficult to obtain 3D face data which has become an obstacle for 3D face recognition to be applied widely.According to it,this dissertation researches the face recognition method based on 3D morphable model.The main work is as follows:Firstly,the inaccurate facial feature points lead to poor-quality model parameters,which makes the unstable shape expression ability of the reconstructed model.This dissertation proposes a reconstruction algorithm to optimize the parameters of 3D morphable model.It locates feature points accurately and then match them with the 3D morphable model to obtain model parameters.Moreover,in order to further improve the quality of the model parameters,it is fused with the parameters obtained based on the regression method to optimize the parameters.Secondly,recognition research based on the reconstruction of the 3D face model is carried out.According to the impact of expression variations on face recognition accuracy,an algorithm combining local keypoints and geodesics curves is proposed.Taking the nose and eyes in a 3D face as a rigid region for keypoints detection,based on which the geometric features are extracted for matching.At the same time,Iso-geodesics extracted are utilized for similarity matching in non-rigid region.Then the matching similarities of the two regions are weighted and fused to obtain the final recognition result.Finally,experiments and comparative analysis are performed on the public data set.Experimental results show that the algorithm proposed in this dissertation could obtain a highprecision face model,and the recognition is robust to expression variations and could effectively identify faces. |
PDF Full Text Request