Efficient closed contour extraction from range image's edge points

This paper presents an improvement over a previous contour closure algorithm. Assuming that edge points are given as input, the proposed approach consists of two steps. Similarly than the previous approach, the minimum spanning tree of a partially connected graph is initially computed. Then, a morphological filter removes noisy links and finally open contours are closed by minimizing a linking cost function. Advantages of the proposed technique lie in the lack of user defined thresholds and non-dependency of edge point density. Experimental results with synthetic and real range images are presented showing encouraging results with uniform and non-uniform edge points' distribution.