HEURISTIC-PROCEDURE FOR SEGMENTATION OF 3-D RANGE IMAGES

A heuristic procedure for the segmentation of 3-D range images is presented. The range image is coarsely segmented into convex, concave, and flat patches using curvature sign labelling at each surface point. The jump edges are then imposed on the coarsely segmented image, resulting in the initial image segmentation. Next, the boundary points between adjacent surface patches are examined for the purpose of merging compatible patches. A boundary point between patches, which is not on a jump edge, has to belong to a crease or to a ridge line. This is detected by the presence of a large local norm of the residuals for an approximating plane on a window surrounding the boundary point. The present method is expected to be faster than many existing methods. Two examples are used in illustration. Comments and conclusions are given. © 1990 Taylor & Francis Group, LLC.