Skip to Main Content
One method of 3D face modeling is constructing a subspace of 3D face image by a set of personspsila 3D face image. It is crucial to register each 3D face image with a reference face image. The current state of the art in image registration is best represented by the well-known iterative closest point (ICP) algorithm and numerous variations on it. Although this method is effective in many domains, it nevertheless suffers from two key limitations: It is not robust to outliers arising from noise and its run time is too long to be applied in real-time systems. To avoid these problems, this paper presents a fast and robust ICP algorithm. The robustness is ensured by improved M-estimator, whose function is defined over a novel residual norm. In addition, because the correspondence matching stage of ICP is usually the one that takes the longest, applications that require ICP to run quickly must choose the matching algorithm with the fastest performance. So we present a fast matching algorithm that has potential application to real-time 3D face model acquisition. Experimental results on 60 personspsila 3D scan face data show the improvements realized in this paper.