【作者】 肖婧；

【导师】 谭阳红；

【作者基本信息】 湖南大学 ， 仪器仪表工程， 2009， 硕士

【摘要】 随着社会的不断发展,手写体字符识别技术被越来越多地应用于生产生活中。尽管研究已经过30多年的时间,但较为通用且识别正确率较高的手写体识别方法仍是人们目前研究的焦点问题。本论文完成的主要工作及创新点体现在如下几个方面:首先,本文针对手写体识别研究的背景、现状及方法、难点及应用前景等进行了一些分析研究。另外,本论文还针对手写体样本预处理的平滑去噪、二值化、图像细化、归一化等步骤的实现原理及方法等做了一定的分析研究。然后,本文在针对手写体识别中常见的粗网格特征、笔划密度特征、投影信号特征和小波变换特征等四种特征提取方法进行分析研究的基础上,提出了两种新的特征提取方法。这两种方法分别为,将粗网格特征与四方向笔划密度特征结合而成的两特征结合特征提取法,以及先提取投影信号然后对其进行小波变换而产生的基于投影信号的小波变换特征提取法。其次,本文在对量子神经网络(QNN)的相关原理及算法等内容进行分析研究的基础上,结合前面提出的两种特征提取方法,提出了两种新的手写体识别方法,其一为基于两特征结合特征提取法与QNN的识别方法,其二为基于投影信号的小波变换及QNN的识别方法。通过采用MNIST手写数字、金融大写汉字、手写英文等三种样本,对两方法进行了Matlab仿真验证,其结果也表明,两方法对于常见的手写体样本类型都能适用,而且识别正确率也比较高。另外,本论文还对两种新方法进行了分析比较。在本论文的最后,对全文进行了分析总结与回顾,并对进一步的研究工作做了展望。更多还原

【Abstract】 With the development of society, the ha n d wr i t t e n-character r e c o g n i t i o n technique has been widely applied to production and daily life. Although the research about the technique has went on 30 years, it is also focus on how to make the handwritten character recognition technique much more common and have higher recognition rate.The innovative ideas and the main work in this paper are as follows.At first, some contents of handwritten-character recognition are instructed, just as the research background, current state of research, research difficulty and application. And then, the principle and implementation of pretreatment processes of handwritten character swatch are presented in this paper. The processes include image noise reduction, binarization, thinning of image, normalizing method.Second, four common methods of feature extraction: rough grid feature extraction, stroke-density feature extraction, the traditional projection characteristic feature extraction and wavelet muhiresolution analysis feature extraction are also researched in this paper. On the basis of these researches, two new methods are introduced. The one is the blend features extraction which is composed of rough grid feature and stroke-density feature extraction, and the other is also a blend features extraction which combines the traditional projection characteristic and wavelet muhiresolution analysis.Third, based on the research of the basic principle and all kinds of learning algorithm of QNN, two new handwritten-character recognition methods are presented in this paper. The one is composed of QNN and the grid feature and stroke-density feature extraction, the other is composed of QNN and a blend features extraction, which combines the traditional projection characteristic and wavelet muhiresolution analysis feature extraction. The simulator results show that common and higher recognition rate for three swatch types include MNIST handwritten number, Chinese character and English letter. Also, two methods are compared and analysed in the paper. At last, it makes the summary and review of the whole paper and the forward research aim.更多还原