Online recognition of Chinese handwritten characters based on the point distribution model

Abstract

High variability is the main difficulty in online handwriting recognition. A wise mechanism should be devised to alleviate this problem. Among others, character deformation is one particular promising approach. Through shape deformation, some of the variations in handwriting is absorbed. In this paper, we present a point distribution model (PDM)-based method for online recognition of handwritten Chinese characters. The PDM technique is a powerful statistical tool that learns from a group of sample shapes to extract their mean shape and analyze their principal variation modes. These variation modes describe the main ways in which the sample shapes tend to deform from their mean shape. In applying the PDM technique to online Chinese character recognition, we treat deformations within the range of two times of the standard deviation of each variation mode as acceptable. For each test character, the variation parameters are strictly limited to be within this range by utilizing projection. The difference of the variation vectors is then used for character classification. We have conducted some experiments to demonstrate the applicability of the proposed approach. The experimental results indicate that the proposed PDMbased approach is superior compared to method 1 in [9] and the SAT approach in [6]. The main reason for better performance should be attributed to the structural deformation capability that the PDM technique owns.