handwritten numeral recognition using flexible matching based on statistics of character variations and shape of contours

摘要

The purpose of this study is to develop a flexible matching method for recognizing handwritten numerals based on the statistics of stroke structures learned from learning samples and shape of character contours. First, this paper describes briefly a method of how to obtain a set of standard character patterns and the ranges of various variations statistically from the given learning samples. In the recognition phase, the matching method deforms each standard pattern flexibly to match with the input character, controlled by the orientations and convexity/concavity of contour points of the input character as well as the internal energy of the standard pattern. Then the standard pattern is dilated inside the input character until it coincides the input. Finally, matching is evaluated in terms of the strain of deformation and the cost of dilation to determine the category of the input character. In this paper we made recognition experiments on samples from two databases and showed that our scheme works well.