3D surface registration using Z-SIFT

We present a Z-SIFT based 3D surface registration algorithm that utilizes the depth information enhanced SIFT features to make initial alignment and the 2D feature weighted Iterative Closest Point (ICP) algorithm to realize accurate registration. The combination of SIFT features and depth information extracts faithful corresponding points between the 2D images and provides good coarse alignment for the 3D surfaces. The 2D feature weighted ICP also outperforms the naive ICP algorithm in terms of speed and accuracy. We use this approach in the context of multiple view alignment for 3D scanners. Experimental results with real objects and human faces indicate the effectiveness of the proposed approach.