Developmental Research for Hand-Written Caracter-Recognition Systems Based on Cellular Networks

本课题的目的是建立一个手写体汉字识别的基本技术，即建立一个能以实用的正确率识别手写体汉字的系统。为了达到这个目的，我们首先开发了一套细胞网络，它能使汉字的大小标准化，提取汉字的笔画，将每个笔画细化为一个单位的粗细，并且以足够高的速度提取字符的全局特征。其次，我们研究了如何利用提取的汉字特征将汉字分成适当数目的组，每个组由在某种意义上相似的汉字组成，并设计了一种算法，使这种分类具有较高的正确率.该算法将以向量形式表示的特征分成若干部分或子向量，对每个子向量进行字符分类，最后统一每个子向量得到的所有分类。该分类是字符识别的第一阶段，并且在字符识别的速度或效率很大程度上取决于该阶段的算法的意义上是非常重要的。与目前使用的各种模式匹配算法相比，我们的算法具有分类速度快、正确率高的特点。最后，我们研制了一套由上述蜂窝网络和一套微机组成的手写体字符识别系统。对手写体汉字识别的各种测试表明，该系统将是建立手写体汉字识别基础技术的一个非常有用的工具。

The purpose of the project is to establish the fundamental technique for recognizing hand-written characters by making such a system that may recognize especially hand-written Chinese characters at a practically useful rate of correctness.To attain the purpose, first we developed a set of cellular networks that enable to normalize characters in size, extract strokes of characters, thin each strok in an unit of thichness, and extract grobal features of characters in a sufficiently high speed. Next we investigated how to use the extracted features of Chinese characters in order to classify them into an appropriate number of groups each one of which consists of similar characters in a sense, and worked out an algorithm which made such a classification possible at a high rate of correctness.The algorithm devides the feature represented in the form of a vector into a number of parts or sub-vectors, classifies characters with every sub-vector and finally unifies all the classifications obtained by each sub-vector. This classification is the first stage of the character recognition and is very important in the sense that the speed or efficiency of the character recognition depends largely on the algotithm in this stage. Our algorithm enables the classification in a high speed and a high rate of correctness compared to various pattern-matching algorithms used up to now.Finally we have developed the hand-written character recognition system that consist of the set of cellular networks mentioned above and a set of micro-computers. It has been made clear by various tests of hand-written Chinese character recognition that the system will be very useful tool to establish the fundamental technique for hand-written character recognition.