Research on Stochastic Models in Character Recognition

该技术对于简单的二值图像的识别，字符识别是其主要的例子，往往使用一定的统计特征。这些方法中的许多可能依赖于个人经验，因此很难为它们找到一个普遍的理论。然后，我们考虑这些技术的分布如下3种类型。以图形表示的分布（实际上，它是通过某些特征向量的多维分布。2.在文档中作为字符符号的分布3.识别系统的分布本研究的目的是通过分析这些分布的各个方面，找到这些技术的一些一般理论。因此，我们开发了字符图像的线型分类技术，线条图形图像的识别技术，并利用各种识别方法的识别顺序分布将几种识别方法结合起来。此外，我们还将这些技术应用于混凝土试件的实测干缩应变的数据预测，并利用一些统计技术对图像扫描仪输入的包含数学表达式的文档进行识别。此外，对于投票方法，从理论的角度进行了进一步的分析，并将其应用于调查问卷数据中固有的人类心理分析。

The technique for the recognition of simple binary images, character recognition is its main example, often uses certain statistical features. Many of these methods may depend on individual experiences, so that it is very hard to find a general theory for them. Then, we consider these techniques by the distributions of following 3 types.1. Distribution as a figure (Actually, it is a multidimensional distribution through some feature vector.)2. Distribtuion as a character symbol in the document3. Distribution of the recognition systemThe aim of this research is to find some general theories underlying these techniques by analyzing various aspects of these distribution. As the result, we have developed the technique of classifying the line types is character images, recoginiton of images of line graphics, and combined several recognition methods using the distribution of recognition order of each recognition method. In addition, we applied these techniques to the data prediction of measured dying shrinkage strain of concrete specimens.And the performance of the recognition of the document including the mathematical expression input by image scanner, using some statistical techniques. In addition, for the voting method, it advanced further analysis from the theoretical standpoint and applied them for the analysis of human mind inherent in the questionnaire data.