摘要
为了提高文字识别的正确率,本文比较详细地探讨了手写文字图象在归一化、细化过程中存在的复杂的随机变形,提出了一些有特色的自适应性较强的预处理方法来解决诸如手写文字图象的总尺寸与其构成的笔划宽度一般不成正比等难题,我们还研究出比较符合人眼视觉效果的快速自适应细化算法(375字/秒),这些处理的结果能很稳定地保持原始文字的结构信息,为识别手写的汉字、字母、数字带来了方便。
The different kinds of random variation of normalized and thinned patterns of handwritten character are discussed in detail in the paper. And, for the purpose of recognition the authors propose some special measures to deal them. A new method of more shapeadaptive thinning is proposed also in the paper. These preprocessing are mainly used for obtaining Well-formed skeletons of binary handwritten character images after normalization and thinning. The best property is preserving the original structure information of input characters to be recognized by OCR systems, which is specially helpful for the recognition of handwritten characters soch as Chhinese characters, English letters and numerals.
出处
《信号处理》
CSCD
1997年第2期132-140,共9页
Journal of Signal Processing
关键词
图象
文字识别
自适应预处理
细化
pattern recognition
handwritten characters
preprocessing
thinning