3D Face Reconstruction Based On Orthogonal Photographs

3D face reconstruction has always been a challenging in the field of computer vision and image processing. It is not only widely used in teleconference and digital movie production, but also used to solve problems such as pose and emotion changes in face recognition, which are very hard to solve under traditional 2D condition.Firstly, this dissertation applies the Active Shape Models (ASM) algorithm to locate the feature points in the frontal and profile view of photos. The initial shape selection, location and the local feature description around the feature points are discussed deeply. In order to process photos with different characteristics of frontal and profile view, this dissertation uses the information of horizontal and vertical directions to select the initial shape and locate it. Then, the profile face is segmented based on the nonlinear color transform the initial shape in profile is selected and located using some located distinct points. An improved method is introduced to describe the local property which combining the local gray property with the local texture property. Secondly, based on these feature points, special 3D face model is obtained by modifying generic Candide-3 model. Then through mapping the special 3D face model with the face texture extracted from the photo, a relative real 3D face model is built, which is close to the person reconstructed. Experimental results show the efficiency of this improved method.The 3D face reconstruction system is developed using VC++ 6.0 and OPENGL under Windows platform. 3D facial images with high senses of reality are obtained based on the photos of frontal and profile view.