笔顺连笔自由的联机手写汉字识别

Online Recognition of Handwritten Chinese Character Free from Stroke Order and Connection

论文针对联机手写汉字识别的笔顺自由、连笔自由问题,在整体DP匹配方法的基础上,提出了一种新的确定笔画对应关系的算法(最小风险算法),并同现有的CubeSearch法、匈牙利算法、近邻优先算法在识别速度、识别率等方面作了比较。该方法克服了现有的一些方法的不足,能高效地同时解决笔顺自由和连笔自由问题,而且由于主要采用全局特征进行识别,对形变和噪音具有很强适应能力。另外,字典可以通过聚类自动生成。另外还制作了原型系统,在对GB2312字符集进行的测试中,取得了较好的效果:在笔顺、连笔完全自由的情况下,单汉字平均识别时间小于0.3s,确定笔画对应关系的正确率达99.1%,识别率达94.5%。

Targeting free stroke order and connection problems on online handwritten Chinese character recognition,we propose Minimum Risk Algorithm（MRA） for optimal stroke correspondence determination based on holistic DP matching method,and compare MRA with Cube Search,Hungarian Algorithm and Nearest First Algorithm in the aspects of recognition speed and rate.This method will not only solve the problems of free stroke order and connection simultaneously but also show a high tolerance for deformation and noise.In addition,the dictionary can be generated automatically.Meanwhile a prototype system is made anti high reeognition performance is proved while testing GB2312 character set.The average recognition time is 0.3s,the rate of stroke correspondence determination reaches 99.1% and the reeognltion rate hits 94.5%.