3D face reconstruction by KLT feature extraction and model consistency match refining and growing

In this paper a new method for the reconstruction of 3D face model is proposed. The method starts with corner extraction and matching using KLT (Kanade-Lucas-Tomasi) approach. Then we apply special algorithm to find and refine erroneous matches. The proposed refining algorithm is a two stage algorithm. In the first stage, we find incorrect matches by sorting displacement vectors. The second stage of the refining process is performed by calculating affine model between two stereo images. In this stage the match points which are not consistent with affine model are considered as incorrect matches. To find match points for the remaining pixels of stereo images, we used match point growing algorithm which uses the refined points and their matches as seed in recursive method. When the match points are obtained, they are used for 3D face model extraction. To remove noise in the reconstructed 3D image, we used adaptive Gaussian filter. Experimental results showed the promise of the algorithm.