Abstract: To solve the problem that high speed and effectiveness is needed for the pre-merger step of online Chinese handwritten text recognition, a multi-layer linear classifier based on minimum risk is provided and realized, which utilizes the geometrical features of strokes and segmented blocks and introduces the scheme of multi-layer combination. Experiments running on real samples show that the classifier reduces the over-segmentation error rate significantly while keeping the under-segmentation error rate in a very low level.

Key words: character segmentation, pre-merger, minimums risk, discriminative learning, multi-layer classifier

摘要： 中文联机手写文本切分识别要求预合并快速高效。为此，设计一个基于最小风险的多层次线性分类器。该分类器根据笔画及切分块的几何特征，采用分层合并的方式完成预合并过程。通过对联机样本进行的实验证明，该分类器保持欠切分错误率在一个较低水平的同时，有效地控制了过切分的错误率。

关键词: 字符切分, 预合并, 最小风险, 鉴别学习, 多层次分类器

CLC Number: 

• TP391.4