Performance improvement of the offline cursive style character recognition system

在这项研究中，我们开发了一个支持系统，为没有专业知识的人阅读古代文献。由于古代文献通常是用印度墨水和草书书写的，这些因素给阅读理解带来了困难。在这个系统中，它显示了汉字的建议到第一到第五，以确定古代文件的字符是哪一个现代字符。针对草书字体形式多样、字与字之间单行连续拼出的特点，提出了一种新的算法。该系统可以通过在数据库中加入“亨泰”字符来应用于许多字符。然后利用像素密度的导数对草书字体进行裁剪。结果，识别率达到了75%。

In this study, we develop a support system for people who do not have the expert knowledge to read ancient documents. Because of the ancient documents are usually written in India ink and in cursive style, these factors make it difficult reading comprehension. In this system, It displays suggestions of characters up to the first to fifth to determine the character of ancient documents is which of the modern character. A novel algorithm is developed cause of cthe cursive style characters have a various forms and spell out in succession with a single line between characters. The system can be applied to a number of characters by including "hentai" characters in the database. Furthermore cursive style characters were cut using the derivatives of the pixel density. As the result, recognition rate has been achieved 75 percent.