Handwritten Text Preprocessing Algorithm Applying SUSAN Principle

In this paper an analysis is presented concerning the applicability of filters and edge/corner detectors using Smallest Univalue Segment Assimilating Nucleus (SUSAN) principle to preprocessing of images containing handwritten text. A new adaptive approach for estimating intensity threshold is given which proves filter and edge/corner efficiency. Different brightness similarity functions are tested for proper finding the Univalue Segment Assimilating Nucleus (USAN) area. A comparison is made with the averaging, Gaussian and median filters to reveal SUSAN filter superiority when filtering different kinds of noise and preserving the original structure of the handwriting.