基于BP神经网络的手写体数字识别分析与研究
详细信息    本馆镜像全文|  推荐本文 |  |   获取CNKI官网全文
摘要
神经网络是一个高度并行的,非线性的,具有很高冗余度的系统,这种系统结构使知识的表达与存储,使模式信息处理过程,都与传统的方法有很大差别。它具有的高度非线性,使我们能表达一些至少是目前尚无法用计算理论表达清楚的外部世界模型;同时,神经网络所具有的自学习,自组织能力使我们能在与外部世界的交互作用下,实现无法用当前的计算理论表达清楚的功能;对于那些无法建立精确数学模型的系统,神经网络有着独特的优势。
     手写体数字识别是多年来的研究热点,也是字符识别中的一个特别问题。手写体数字识别在特定的环境下应用特别广泛。当涉及到数字识别时,人们往往要求识别器有很高的识别可靠性,特别是有关金额的数字识别时,因此,针对这类问题的处理系统设计的关键环节之一,就是设计出高可靠性和高识别率的手写体数字识别方法。然而可以说还没有哪个手写体数字识别器达到完美的识别效果。
     在过去的数十年中,研究者们提出了许许多多的识别方法。按使用的特征不同,这些方法可以分为两类:基于结构特征的方法和基于统计特征的方法。
     手写体数字识别属于模式识别的范畴,是模式识别中的一个具体问题,因而,对于手写体数字识别有着不同于其它模式识别的特殊要求。手写体数字识别除了要求识别精度高和工作可靠外,还要求其识别效率高,识别数度快。
     由于手写体数字识别难于建立精确的数学模型,所以本文采用BP神经网络对这一问题进行处理。神经网络模式识别的一个关键步骤是预处理和特征提取,对于手写体数字识别,本文采用了一种基于字符边缘轮廓的特征提取方法,并用程序实现了这一特征提取过程。本文还训练出了具体的用于数字识别的神经网络,用程序模拟了整个识别过程。本文采用了两级神经网络结构。
     通过测试,本识别系统对于较规范的手写体数字的识别达到了很好的识别效果。
The neural networks is a collateral, nonlinear and redundancy system, and it makes the express, memory and treat of information varied from routine. It's the non-linearity that makes us could express the model that can't be expressed clearly by compute theory at least now; and it's the ability of self-study and self-organization that makes us can express those clearly by acting with exterior world, which can't be expressed by compute theory. The neural networks has unique predominance for system that can't make accurate mathematics model.
    Handwritten numeral recognition is a hotspot of study for years, and is some especial issue of character recognition. Handwritten numeral recognition is applied broadly in given environment. When come down to numeral recognition, the emphases people think is its dependability, especially refer to money-digit recognition. So one of the key steps for these questions is designing a high-dependability and high-accuracy handwritten numeral recognition system. But there are no system can achieve so good recognition effect.
    In the past years, researchers had put forward many recognition ways, which can be carved up to structure character oriented ways and statistic character oriented ways.
    Handwritten numeral recognition belongs to pattern recognition, and is a factual question of pattern recognition, so it has some special request. The handwritten numeral recognition system must be efficient and fast besides dependable and accuracy.
    It's difficult to make accurate mathematics model for handwritten numeral recognition, so BP neural networks is used here. The key steps of neural networks pattern recognition are preprocessing and feature subset selection. In this paper, algorithm of feature subset selection basing on the edge-outline
    
    
    of characters has been adopted in handwritten numeral recognition, and the process of feature subset selection had been realized in program. Neural networks for handwritten numeral recognition had been trained in this paper, and its structure is two classes.
    Recognition system in this paper has achieved a good rate of recognition in random handwritten numeral by test.
引文
[1] 黄德双,神经网络模式泌别系统理论,电子工业出版社,1996.5。
    [2] 李士勇,模糊控制·神经控制和智能控制沦,哈尔滨工业大学出版社,1998.9。
    [3] 边肇祺,模式识别,清华大学出版社,1988.6。
    [4] 黄凤岗、宋克欧,模式识别,哈尔滨工程大学出版社,1998.3。
    [5] 殷勤业、杨宗凯、谈正,模式识别与神经网络,机械工业出版社,1992.9。
    [6] 焦李成,神经网络的应用与实现,西安电子科技大学出版社,1993.6。
    [7] Mori S.,Suen C.Y.,Yamamoto K.,Historical Review of OCR Research and Development, Proc.IEEE,1992,80(7),1029-1057.
    [8] Govindan V.K.,Shivaprasad A.P., Character Recognition-A Review,Pattern Recognition,1990,23(7):671-683.
    [9] C.Y. Suen, M.Bertoh,and S.Mori, Automatic recognition of hand-printed characters-The state of the art, Proc.IEEE,vol.68,pp.469-487,1980.
    [10] C.Downton and S.Impedovo eds., Progress in Handwriting Recognition, World Scientific Publishing Co.Pte.Ltd., 1997.
    [11] S.Impedovo, P.S.P. Wang and H.Bunke(eds.), Automatic Bankcheck Processing, World Scientific Publishing Co.Pte.Ltd., 1997.
    [12] 黄德双、马颂德,“神经网络模式识别的基本特征和基本理论”,中国博士后十周年庆祝学术大会论文集,北京,1995.10。
    [13] 周继成、周青山、韩飘扬,人工神经网络:第六代计算机的实现,科学普及出版社,1991。
    [14] 傅京孙,模式识别应用,北京大学出版社,1984。
    [15] 于常友、陈伟基,自动化科学和技术,电子工业出版社,2000。
    [16] 丛爽,面向MATLAB工具箱的神经网络理论与应用,中国科学技术大学出版社,1998.11。
    [17] 何玉彬、李新忠,神经网络控制技术及其应用,科学出版社,2000.11。
    [18] 郑君里、杨行竣,人工神经网络,高等教育出版社,1992。
    [19] 蔡元明,神经网络识别手写体数字预处理后样本空间凸集性研究,中国科学院半导体
    
    所硕士学位论文,1995.6。
    [20] 张立明,人工神经网络的模型及其应用,复旦大学出版社,1992。
    [21] 王新贵、刘建胜、居琰、汪同庆、彭健、杨波,“有效行”特征对手写体字符的识别,电子科技大学学报,2001.6第3期。
    [22] 王雯、施鹏飞,多层分组神经网络的手写体数字识别,上海交通大学学报,1998.9第9期。
    [23] 胡钟山、娄震、杨静宇、刘克、孙靖爽,基于多分类器组合的手写体数字识别,计算机学报,1999.4 第4期。
    [24] 王伟、胜立东,基于级连分组BP网络的高精度手写数字识别,中文信息学报,1995年第14卷第2期。
    [25] 黄心晔、王茂祥、富煜清、陆佶人,基于结构分析的手写体数字识别算法,电子工程师,1999年第11期。
    [26] 胡钟山、娄震、杨静宇、刘克、孙靖夷,基于轮廓分段特征的手写体阿拉伯数字识别,计算机学报,1999.10第10期。
    [27] 刘来元、李炳成、马颂德,基于曲线矩的手写体数字识别,模式识别与人工智能,1995.6第2期。
    [28] 施善昌,自动识别原理与应用,人民邮电出版社,1989。
    [29] 陈鸣华、阎平凡,基于数字形态学的手写体数字识别方法,自动化学报,1989.5第3期。
    [30] 张保轩、房世晖,基于外形特征的手写体数字识别,山东通信技术,1995年第1期。
    [31] 刘勇、赵斌、夏绍玮,模糊超椭球分类算法及其在无约束手写体数字识别中的应用,清华大学学报,2000年第9期。
    [32] 谢光毅、钟义信,神经网络用于手写体数字识别,模式识别与人工智能,1994.12第4期。
    [33] 朱学芳,手写数字识别实验系统的研究,南京大学学报,1996.1第1期。
    [34] 候继红、徐军,手写体数字识别技术的研究,电子计算机与外围设备,第23卷第5期。
    [35] 朱江、宣国荣,一种基于骨架特征顺序编码的脱机手写体数字识别方法,小型微型计算机系统,2001.8第8期。
    [36] 吴雪菁、施鹏飞,质心层次特征的无约束手写体数字识别,上海交通大学学报,1998.9第9期。
    
    
    [37] Nadal C., Legault R., Suen C.Y., Complementary algorithms for the recognition of totally unconstrained handwritten numerals, Proc. 10th Int. Conf.Pattern Recognition,June 1990,Vol.A,pp.434-449.
    [38] C.Y.Suen,et al.,Recognition of totally unconstrained handwritten numerals based on the concept of multiple experts,Proc.Int.Workshop on Frontiers in Handwritten Recognition, Montreal, Canada, pp. 131-143,1990.
    [39] Avi-Itzhak H.I., Diep T.A. and GarlandH., 1995, "High Accuracy Optical Character Recognition Using NeuralNetworks with Centroid Dithering",IEEE Trans. on Pattern Analysis andMachine Intelligence, Vol. 17, No.2, pp.218-223.
    [40] C. Kaynak, Methods of Combining Multiple Classifiers and Their Applications to Handwritten Digit Recognition, MSc Thesis, Institute of Graduate Studies in Science and Engineering, Bogazici University, 1995.
    [41] L. Davis. Handbook on Genetic Algorithms. Van Nostrand Reinhold, 1991.
    [42] J.Yang and V.Honavar. Feature subset selection using a genetic algorithm. IEEE Intelligent Systems, 13(1):44-49, 1998.
    [43] McCulloch, W.S. and Pitts, W., "A logical calculus of the ideas immanent in nervous activity," Bulletin of Mathematical Biophysics, vol. 5, pp. 115-133, 1943.
    [44] Rosenblatt, F., Principies of Neurodynamics, New York: Spartan, 1962.
    [45] G.L. Martin, J.A. Pittman, "Recognizing Hand-Printed Letters and Digits Using Backpropagation Learning", Neural Computation 3, 258-267, 1991.
    [46] Zhengquan H. and Siyal M.Y., 1998, "Recognition of Transformed English Letter with Modified Higher-Order Neural Networks", Electronics Letter,Vol. 34, No.2, pp. 2415-2416.
    [47] 易克初、姜建新、胡征,用于语音识别的神经网络研究,西安电子科技大学学报,第18卷,增刊,87—95,1991。
    [48] 靳蕃、范俊波、谭永东,神经网络与神经计算机,西南交通大学出版社,1991。
    [49] 张析中,汉字识别技术,清华大学出版社、广西科学技术出版社,1992。