This paper presented a novel method to select the edge scale automatically, which could be used to split complicated interleaving edges into more meaningful edge segments that could be classified easily according to their features. First, edge extreme paths in Gaussian scale space were searched to obtain the characteristic scale for each edge point, by calculating the maximum distance the edge point travels from one scale level to adjacent one and analyzing the four forms of extreme path evolution. Then, interleaving edges were split into pieces with different scales according to their scale histogram. These edge pieces connecting with each other were combined through extreme path in scale space. Experimental results show that the proposed method is effective in edge segmentation based on characteristic scales and saliency of edge.

Key words：