SUSAN edge detector reinterpreted, simplified and modified
Ewaryst Rafajłowicz · 2007
Our aim is to provide a new interpretation of the SUSAN edge detector and to propose its modified version, which is still simple, robust against errors and provides thinner edges, without further efforts on additional thinning. The new interpretation uses the idea of vertically weighted regression and in its simplest form it leads to interpreting USAN in terms of a box sliding on the surface of an image. This way of interpreting the SUSAN detector suggests hints on tuning its parameters. It also reveals why and to what extent this detector is robust against errors. Guided by this interpretation we also propose simple modifications of the SUSAN algorithm, which provide thinner edges without any additional thinning procedure, keeping other advantages unchanged.