NEW CHARACTER RECOGNITION SYSTEM BASED ON STRUCTURAL INFORMATION OF CHARACTERS

本课题的研究目的是解决将用于印刷字符识别的大字典分类为大量小字典的问题，从而彻底解决该问题，使字符识别速度更快。我们通过开发一种新的技术和算法，利用一些适合分类的字符结构参数，将待识别的字符集划分为两组，并具有足够高的准确率，从而达到了这一目的。这些结构参数主要与汉字的结构复杂性有关，特别是日文特有的汉字，例如“假名”和“汉字”，以及汉字直方图的配置。另一方面，通过在两组字符的边界附近引入称为灰色区域的区域来定义，并允许灰色区域内存在的每个字符都属于两组字符，从而获得极高的分类精度。当然，我们在研究项目中开发的技术和算法将使基于模式匹配的字符识别系统通过将一个大字典分类为5个小字典进行识别，使其速度至少比通常的识别系统快5倍。

The purpose of the research project is to solve the problem to classify a large dictionary for printed character recognition into a large number of small dictionaries, which results in solving drastically the problem to make the character recognition faster. We have attained the purpose through the development of a new technique and an algorithm to use some structural parameters of characters suitable for classification of the set of characters to be recognized into twu groups with sufficiently high accuracy.These structural parameters are mainly concerned with the structural complexity of characters especially peculiar to Japanese characters, e.g., 'Kana' and 'Kanji' and also with the configuration of histograms of characters. On the other hand, the extreamely high accuracy of classification is obtained by introducing the region called as a gray zone to be defined in the neighbor of the boundary of two groups of characters and to allow every character existing in the gray zone to belog to both groups.It is certainly expected that our technique and algorithm developed in the resarch project will make such character recognition systems as based on pattern matching at least five times as fast as usual ones by classifying a large dictionary for recognition into may small dictionaries.