Abstract: A new stroke tracking algorithm which based on the key points of stroke segmentation and least-squares fitting is proposed according to a current situation that directly tracking of complex Latin cursive strokes is still a difficult problem. Each kind of template strokes’ key points is acquired from figure domain and time domain, and is used for extracting the stroke and acquired fitting-curve’s coefficient vector template by least squares. The key points from real writing and fitting curve（reconstructed by coefficient vector） are used for computing the similarity. Conic curve, sinusoid and line have been chosen as standard fitting curve. The experiments show that the method is objective and robust, besides, it can be applied to more than two ligaturing-letters stroke tracking and other kinds of Latin writing language.

Key words: ligaturing-letters, stroke tracking, extract of key points, curve fitting, coefficient vector

摘要： 针对目前笔画跟踪算法不能直接跟踪分析拉丁文字母连写笔画的问题，以英文二连字母笔画为例，提出基于关键点进行笔段分割与最小二乘法拟合的字母连写笔画跟踪方法。在图域、时域获取模板笔画各类笔迹关键点向量，依据关键点提取笔段，利用最小二乘法将笔段转换成拟合曲线的系数向量模板；将实写笔画的各类关键点与系数向量重构的拟合曲线进行贴近度分析从而实现笔迹跟踪。范类拟合曲线选择二次曲线、正弦曲线及直线。实验与分析结果表明，该方法对字母连写笔画跟踪效果具有较强客观性、鲁棒性，其原理同样适用于多于二字母的连写笔画跟踪，也可推广到其他拉丁文文种。

关键词: 字母连写, 笔迹跟踪, 关键点提取, 曲线拟合, 系数向量