Reliability Pyramid Based Dense Corresponding between Facial Texture Images

Dense corresponding between facial texture images is the core step of the morphable model based 3D face reconstruction. In this paper, we propose an original dense corresponding algorithm by means of the reliability pyramid, which utilizes the output of highly accurate sparse matching algorithm as the initial estimation of the iterative Lucas-Kanade optical flow. Furthermore, two constraints for facial texture corresponding and a rapid 3D face reconstruction strategy are also briefly introduced. We conduct experiments on the BJUT-3D Face Database, proving the above-mentioned theories to be powerful.