Toward developing a method of high accuracy character recognition for poorly handwritten Kanii characters.

致力于开发一种针对手写不佳的 Kanii 字符进行高精度字符识别的方法。

基本信息

批准号：
12480093
负责人：
WATANABE Yoichi
金额：
$ 5.31万
依托单位：
Yamagata University
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (B)
财政年份：
2000
资助国家：
日本
起止时间：
2000 至 2002
项目状态：
已结题

项目摘要

We developed a quantification method to assess the legibility of handwritten Kanji characters. The index of legibility deriving from this method corresponded to the results of psychological experiments. Sample characters were 528 handwritten Kanji characters selected from signatures written by 280 students. Other 33 participants were asked to classify these characters into three levels of legibility: Well ・ Ordinary ・ Poor. Then a quantification method of legibility of handwritten Kanji characters was introduced. This method adopted some variables corresponding to the property of human cognition: global features and local features of characters. With this method, we could predict human decision about legibility of handwritten Kanji characters at about 90% precision level. For example, the index of this method showed quantitatively that there were some handwritten characters having higher legibility than block letters. Next, we analyzed patterns of the eye movement during the character recognition task. The stimuli were block-letter, well-hand written, and poorly-handwritten Kanji characters. The percentage of correct recognition was highest with the block-letter, second highest with the well-hand written, and lowest with the poorly-handwritten characters. There was a tendency that the number of fixation went higher in accordance with the increase of the poor handwriting index, which the authors had invented. In conclusion, the validity of this index was confirmed. We believe this index is an useful measure to research the high accuracy recognition system of handwritten Kanji characters.

我们开发了一种数量方法来评估手写汉字字符的可读性。从这种方法得出的可读性指数对应于心理实验的结果。样本字符是从280名学生写的签名中选择的528个手写汉字字符。其他33名参与者被要求将这些角色分为三个级别的知识级别：良好的・普通・贫穷。然后引入了手写汉字字符的可读性的量化方法。该方法采用了一些与人类认知属性相对应的变量：字符的全局特征和本地特征。通过这种方法，我们可以预测人类关于手写汉字角色可读性的决策，其精度级别约为90％。例如，该方法的索引数量地表明，有些手写字符比块字母具有更高的透明度。接下来，我们在角色识别任务中分析了眼睛运动的模式。刺激是字母，手写良好的汉字角色。正确识别的百分比在封锁通信的情况下最高，第二高，手写良好，而手写不佳的角色则最低。有一种趋势是，根据作者发明的不良手写指数的增加，固定数量的数量越高。总之，该指数的有效性得到了证实。我们认为，该指数是研究手写汉字角色的高精度识别系统的有用度量。