Algebraic error analysis for surface curvatures and segmentation of 3-D range images

摘要

This paper is divided into two parts. In the first part an algebraic error analysis for the calculated surface Gaussian and mean curvatures of 3-D range images is presented. The error analysis results are used to illustrate the effect of noise on the segmentation of range images using the curvature 8-sign label scheme. In the second part a segmentation procedure for 3-D range images based on conclusions deducted from the first part is proposed. In this procedure, jump edges are first obtained. Then a curvature 3-sign label scheme is used for the segmentation. A simple method is used to check the boundary points between segmentation patches. If a boundary point lies on a planar region it is deleted. In the planarity test we use some of the intermediate results from the first part of the paper. Experimental results and comments are given.