Feature Points Detection And Matching Methods Study Based On Multi-view Face Images

3D Face Reconstruction is a hot topic in the field of computer vision. Since the stereoscopic effect of three-dimensional face model is remarkable, it is widely used in film and television, medical, gaming, advanced face recognition technology and somatosensory interactive technology to three-dimensional development of data processing direction.3D Face Reconstruction based on 2D face image is the process of reducing 2D face image to 3D face model by a series of scientific computing. In the 3D Face Reconstruction process, detection and matching facial feature points influences modeling results greatly, it is an important part of 3D Face Reconstruction.based on a review of existing 3D Face Reconstruction, for detection and matching facial feature points, this paper carried out the following work:(1)For classic feature point detection method can not adapt to different perspectives facial feature detection problem, this paper proposes a face feature point detection method based on edge curvature feature. This method gets the edge enhancement feature by traversing the edge points detected by the Canny operator, and calculating the curvature of each point. The feature points are obtained by choosing appropriate threshold values. Experimental results show the proposed method can detect a large number of feature points of facial features and have good adaptability to multi-view images.(2)For the problem that there are missing points pairs of the BRIEF characterization method, this paper designs a feature points matching method based on BRIEF combination and Euclidean distance. This method discribes the point X with BRIEF descriptors, than gets the Euclidean distance intervals by the method of Hamming distance ratio constraint, thereby obtaining a candidate matching point set, and than determines a matching point of X. Experimental results show that the matching feature point pair in the presented method is less mismatching.(3)Deforming the estimated 3D coordinates of the matched feature points, the 3D face model is reconstructed on the existing general face model (Candide-3). A 3D face reconstruction prototype system is realized based on the presented algorithm.