Dynamic threshold determination by local and global edge evaluation

In this paper we consider two methods for automatically determining values for thresholding edge maps. In contrast to most other related work they are based on the figural rather than statistical properties of the edges. The first approach applies a local edge evaluation measure based on edge continuity and edge thinness to determine the threshold on edge magnitude. The second approach is more global and considers complete connected edge curves. The curves are mapped onto an edge curve length/average magnitude feature space, and a robust technique is developed to partition this feature space into true and false edge regions. A quantitative assessment of the results on synthetic data shows that the global method performs better than the local method. Furthermore, a qualitative assessment of its application to a variety of real images shows that it reliably produces good results.