This paper presents a thinning algorithm which involves the identification of local features like line segments, tips, and junctions by the use of rough set theory and Euler number calculation within a rectangular window operator. It handles image objects which are already presented in binary image format. The resulting skeleton preserves the topological properties of the original shape in the form of a graph with nodes presenting the local features and arcs for the adjacency relations. The algorithm offers two distinctive advantages in terms of conceptual simplicity and computational effort. In general, it generates the skeleton in one pass plus an auxiliary scan confirming the identification of some features in the ambiguous segment regions. The algorithm provides skeletons of good quality for character objects, and so can be used later for syntactic recognition of the alphabets. Excellent results have also been obtained for general shaped objects.
Joe C. H. Poon,
Gary M. T. Man,
Tony C. F. Ng,
"New thinning algorithm using rough-set theory", Proc. SPIE 1906, Character Recognition Technologies, (14 April 1993); doi: 10.1117/12.143636; https://doi.org/10.1117/12.143636