Topological stacking grayscale thinning for edge detection and real-time applications

In this paper, it is introduced a low computational cost technique for grayscale thinning, being also capable of thinning binary images. The technique uses an iterative procedure that constructs a topological stack of eroded and smoothed images, on which the neighborhood related local maxima pixels are morphologically thinned. For grayscale intensity saturated edges, or binary edges, the topological stacking successfully turns the maximally inscribed pixels into local maxima. This procedure guarantees one pixel wide line-drawing for the skeletons and removes distortions. Tests on Berkeley Segmentation Dataset and Benchmark showed the proposed thinning achieved better scores than its competitors regarding post-processing thinning for edge detection, being suitable for real-time applications.