Study On The Method Of 3D Face Reconstruction Based On Binocular Vision

3D(three-dimensional) face model as a very interesting and particularity part of human characteristics, has been a wide range of applications in many fields, such as 3D face identification, 3D games, video conferencing and medical. As its complexity and variability, generation of one lifelike human face model is a very challenging task, by now, most of the techniques are based on the use of expensive equipments or demanding special situations. With the development of computer vision technique and camera calibration technique, 3D face reconstruction based on stereo vision as a cheap, easy to use, high automation method has been attracted more and more attention in recent years. In this paper, considering the application in face recognition, a method of 3D face reconstruction based on binocular vision is studied to build a convenient and effective system of 3D face reconstruction, which include stereo matching, region growing techniques, camera calibration, image correction, reconstruction region extraction and 3D reconstruction.1) After analyizng and investigating current camera calibration methods, Zhan’s method is used to calibrate the two cameras, and their exterior and inner parameters are calculated. Secondly, the parameters are used to rectify stereo images to align their epipolar lines and compensate for image distortions. The experimental result shows that the proposed method is effective.2) Face detection methods based on Adaboost algorithm and based on the model of color skin are both investigated. After the face position fixed, morphologic method is used to reduce the region of face recostruction. Then a new coordinate restrain method is proposed to make sure that the epipolar-lines have not been affected in the stage of region extracted.3) A new method based on tereo macthing algorithm combined with the region growing strategy is proposed. A local stereo matching algorithms—Realtime stereo matching algorithm is improved, and region growing strategies based on seed points vision difference is used for complemention. First, the new Realtime stereo matching algoritm is used to calculate some pixel as seed points. Then region growing processing based on WTA algorithm macthing along horizontal scan-lines under multi-constraints to reduce the system takes. In order to improve the accuracy of the macthing points, the seed points are sorted based on Realtime threshold and reliability before region growing start.4) For the purpose of reconstruction 3D face models in real time and proving practicability of these algorithms proposed upon, a software and hardware platform are designed. Then texture rejection method base on OpenGL is used after the face model is built. Finally, by programming based on Open CV, Visual 2008 and OpenGL together, the experimental result shows that the proposed method can generate a smooth and vivid 3D face model.