| Face, as the most expressive part of human being and with its characteristics of individualizationand diversification, is the most complicated study subject in the field of computer vision. With thedevelopment of computer technology,3D face reconstruction plays more and more important role inface recognition, production of film and TV, sentiment analysis, animation games and medicine, and ithas gradually become a research hotspot in the fields of computer graphics, computer vision andpattern recognition.3D face reconstruction based on single image just needs a face frontal imagewhich is acquired easily, and aroused extensive attention of researchers. This thesis has done someresearch in this field, and the main work is summarized as follows.First of all, the3D face data which are acquired by Cyber-Ware laser scanner is so large andirregular. So it is needed to process the correspondence between faces. The normalization of3D datais the important prerequisite of3D face reconstruction and recognition. In this paper the3D face datais normalized by2D template-based alignment algorithm, and each face is comprised by23676vertexes.Secondly, this thesis comes up with a calculation algorithm of3D face reconstruction based onfeature division. Each face is divided into four pitches (eyebrow pitch, eyes pitch, nose pitch, mouthpitch). This method firstly uses sparse morphable model to obtain the3D data of every2D facedivision, and then uses radial basis function to remedy the3D data. Experimental results show that themethod improves the reconstruction accuracy and has good efficiency.Thirdly, in this thesis3D face reconstruction and3D face recognition are combined to study. Andthe algorithms of3D face recognition are summarized. On the base of the reconstruction algorithm3Dface recognition is proposed, and reaches a high recognition rate.Finally, this paper carried out experimental verification using Matlab and VC++andimplemented the3D face reconstruction and recognition. |
PDF Full Text Request