手写体汉字在特征空间的可视化分析  

The Visualization Analysis of Handwritten Chinese Characters in Their Feature Space

作　　者：陈津颖[1] 金奕江[1] 马少平[1] 

机构地区：[1]清华大学计算机系智能技术与系统国家重点实验室,北京100084

出　　处：《中文信息学报》2000年第5期42-48,共7页Journal of Chinese Information Processing

基　　金：国家重点基础研究! (G19980 30 5 0 9);自然科学基金 (6 96 75 0 0 4;6 9836 0 40 );86 3高技术资助!项目 (86 3- 30 6-ZD0 3- 0

摘　　要：手写体汉字特征一般在几百维以上 ,在这样的高维空间中 ,汉字样本是如何分布的 ?本文从可视化的角度对这一问题进行了探讨。论文首先给出了所选用的汉字特征的定义 ,然后对一些具有代表性的汉字实例 ,从K L变换法、线性投影法和非线性投影法三个方面 ,对汉字在特征空间的分布问题进行了可视化分析 ,结果表明 ,可视化分析可以帮助人们了解汉字在特征空间的分布情况 ,对改进识别器的性能具有指导意义。The feature vectors of handwritten Chinese characters are often more than several hundred.In such a high dimension space,what's the distribution of Chinese characters? The paper discussed this problem through visualization.At the beginning,we gave a definition of the Chinese characters' features that are used in this paper.Then by using K L Transformation,Linear Projection and Nonlinear Projection,we made a visualization analysis to the distribution of some typical samples of Chinese characters in their feature space.The results showed that visualization analysis could help us understand the distribution of Chinese characters in their feature space,thus being instructional to improving the performance of classifier.

关 键 词：汉字识别 可视化分析 特征空间 手写体汉字 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]