中文摘要：

在识别矢量笔迹文本时,不同类型单字需要采用不同识别器,确定详细类别是单字识别的前提。对实际中文矢量笔迹文本中单字进行汉字、标点、数字、字母和单词的详细分类,提出了自身和相对（包括近邻和同行）特征,选用决策树、逻辑模型树、贝叶斯网络和支持向量机四种分类器。针对大量实际数据,测试和比较了多种特征和分类器的性能。实验表明,近邻单字的组合特征具有较好的分类能力,支持向量机对各种单字均有较好分类性能。

英文摘要：

Different types of characters from Chinese ink texts are recognized before they are need to different recognizers. Thus it is prerequisite to identify writing characters＇ detailed categories for improving their recognition. This paper aimed to classify writing characters into Chinese character, punctuation, digit, number, as well as English letter and word. Extracted each writing character＇ self and relative features, and applied representative classifiers, such as decision trees, logistic model trees, Bayesian network and SVM. Features and classifiers were evaluated with many real-life Chinese ink texts. Experimental results show that relative features are more powerful and SVM is the most efficient classifier for each type of writing characters.