Segmentation of On-line Cursive Handwritten Chinese Word Based on Stroke Speed Feature and Stroke Vector Feature

On-line handwritten Chinese word recognition has recently become an important research topic in the filed of computer vision. However, the segmentation of cursive Chinese word is still an unsolved problem. In this paper, two new features, stroke speed feature and stroke vector feature, are proposed for the segmentation of on-line handwritten Chinese word. Analysis and experiments show that both of the features are easy to implement, with low computation complexity and encouraging correct segmentation accuracy. Furthermore, the stroke vector feature outperforms traditional histogram method and we found it is especially suitable for the segmentation of cursive handwritten word where two characters touch each other or overlap.