权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Study on Recognition Mechanism of Transformation Character Pattern by Neural Network

神经网络变换字符模式识别机制研究

基本信息

批准号：
09650434
负责人：
SUZAKI Kenichi
金额：
$ 2.18万
依托单位：
Fukuoka Institute if Technology
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (C)
财政年份：
1997
资助国家：
日本
起止时间：
1997 至 1999
项目状态：
已结题

项目摘要

The person is the good at recognition of the character, the sound, and the image pattern. However, even if the latest image processing technology is introduced, the pattern recognition is difficult. The study on the pattern recognition by the neural network is actively done from such a background. This study is a summary of the recognition mechanism of the character pattern seen when the aspect in the space three dimensions moves. If the aspect by which the character is seen is moved, the character receives geometrical conversion. If the aspect approaches the center of the character plane, the character pattern receives the perspective. At present, the neural network by Which such a transformation character can be recognized is not proposed. This study introduces the copy learning method into the neural network, and proposes the recognition mechanism which can recognize the transformation character. The summary of this study is as follows.(1) The relation between the input pattern resolution and the character recognition rate in the aspect movement was clarified through the experiment. This study plays elucidation of complementary of the aspect movement and the character recognition the role.(2) This study proposes a copy-learning model which can recognize standard characters from superimposed characters by only learning the standard characters. This study is scheduled development into the separation and the recognition of the superimposed character of the dirt character, the picture, and the character for the future.

善于识别文字、声音和图像模式的人。然而，即使引入最新的图像处理技术，模式识别也是困难的。神经网络模式识别的研究就是在这样的背景下活跃起来的。本研究总结了物体在三维空间运动时所看到的字符图形的识别机制。如果看到角色的方面被移动，则角色接受几何变换。如果纵横比接近字符平面的中心，则字符图案接收透视。目前，还没有提出能够识别这种变换特征的神经网络。本研究将复制学习方法引入到神经网络中，提出了一种能够识别变换字符的识别机制。本研究的主要结论如下：(1)通过实验，明确了输入模式分辨率与方位运动中字符识别率之间的关系。提出了一个复制学习模型，该模型通过只学习标准字就可以识别出标准字和重叠字。本研究是对土字、图文、字的叠加字的分离与识别，是今后研究的预定发展方向。