摘 要
目前,模式识别领域在日常生活中的应用已经越来越广泛,比如人脸、指纹识别,字符识别,车牌识别。所以,对字符识别进行学习与研究是非常有必要的。
本课题为数字字符识别模拟演示系统。主要是利用正态分布下的最小错误率Bayes方法和最小风险Bayes方法,来实现手写数字从0到9的识别。该系统首先是实现模拟手写数字;然后利用5*5的模板提取出样品的特征,采用模板可以使同一形状、不同大小的样品得到归一化的特征提取,所以有能力对同一形状、不同大小的样品视为同类;最后结合Bayes决策进行判别。使用最小错误率Bayes方法,在判别过程中能使错误率达到最小,即使错分类出现的可能性最小,而最小风险Bayes方法,在判别过程中可以使风险达到最小,减少危害大的错分类情况。
本设计是利用Visual C++ 6.0实现的,实验证明,该系统对于模拟手写的数字基本上能正确识别,但是对于手写不规范的数字会存在错判的情况,这跟样品库的有限有关。
关键词: 最小错误,最小风险,特征选择,模拟手写,MFC
ABSTRACT
Now,the application of Pattern Recognition is more and more popular in our daily life,for example the man’s Face、Fingerprint recognition, Character recognition License plate recognition.
This project is talking about the digital character recognition simulation demonstration system.It mainly use Least error of Bayes method and Least risk of Bayes method under the Normal Distribution to realize handwritten figures recognition from 0 to 9. First, handwritten digitals should be simulated;then, extract the sample characteristics by useing 5*5 templates,using templates can get feature what not very difference from samples what have the same appearance and different size,so the samples what have the same appearance and different size would be divided to similar
;at last,combine with Bayes method to recognition , this two kinds of methods avoided a lot of classification errors and attaining the wrong probability to be smallest.
This design realized by using Visual C++ 6.0, experimental evidence , this system for any handwritten figures could correctly identify.But for the relatively small handwritten figures ,miscalculated the probability will be greater,the reason is that the sample is limited,usually,the number of samples to the number of features from five to ten times.
Key :Least error,Least risk, Feature selection,Handwritten simulation,
MFC