Matching Algorithm Of Face Reconstruction Based On Binocular Vision

3D face reconstruction based on binocular vision involves many fields, such as applied mathematics, image processing, computer vision. It has become the hot issue both in applied mathematics and high technology. It has potential applicable prospect in face recognition and facial animation, etc.In face reconstruction based on binocular vision, stereo matching is a key step, its accuracy directly affects the result of face reconstruction. In this thesis, we propose the improved matching algorithm based on region growing after camera calibration and epipolar rectification, the main contributions are summarized as follows:Firstly, we introduce the principle of binocular stereo vision and the camera calibration methods. For the target images, we locate the corner points by improved corner detection algorithm based on Harris operator, and then we calibrate the camera.Secondly, for the face images, we use the improved matching algorithm based on window to get seed points, we spread the region growing rules to up and down or so four directions, and we use parabola fitting algorithm to get sub-pixel level disparity.Thirdly, for the disparity we obtain, we remove the wrong disparity by symmetry and continuity constraints, and then we fill the disparity by surface fitting which proposed in this thesis.Finally, we analyze the experimental result of stereo matching, then we discuss and summarize the difficult points of stereo matching which need to be considered in the future of stereo matching.Using the matching algorithm proposed in this thesis, we have got satisfying disparity map and reconstructed the points cloud of3D face.