×

Orientation-independent recognition of handwritten characters with integral invariants. (English) Zbl 1194.68202

Suzuki, Masakazu (ed.) et al., The joint conference of ASCM 2009 and MACIS 2009. 9th international conference on Asian symposium on computer mathematics and 3rd international conference on mathematical aspects of computer and information sciences, Fukuoka, Japan, December 14–17, 2009. Selected papers. Fukuoka: Kyushu University, Faculty of Mathematics. COE Lecture Note 22, 252-261 (2009).
Summary: We present an approach to recognize handwritten characters independently of their orientation. The method is based on the theory of integral invariants and yields good results in classifying rotated samples. We propose two recognition techniques taking advantage of integral invariants up to second order. Truncated Legendre-Sobolev series are used to represent the invariant functions and recognition is based on proximity to the local convex hulls of known classes. We compare performance of these new methods with another widely used recognition method based on geometric moment invariants. The results obtained indicate that integral invariants give better recognition rates with less computation, confirming they are suitable for classification of rotated handwritten characters in a pen-based environment.
For the entire collection see [Zbl 1181.68015].

MSC:

68T10 Pattern recognition, speech recognition
PDFBibTeX XMLCite