基于神经网络手写体数字识别系统的设计与实现.doc

摘要:数字识别就是通过计算机用数学技术方法来研究模式的自动处理和识别。随着计算机技术的发展，人类对模式识别技术提出了更高的要求。特别是对于大量己有的印刷资料和手稿，计算机自动识别输入己成为必须研究的课题，所以数字识别在文献检索、办公自动化、邮政系统、银行票据处理等方面有着广阔的应用前景。

　　对手写数字进行识别，首先将汉字图像进行处理，抽取主要表达特征并将特征与数字的代码存储在计算机中，这一过程叫做“训练”。识别过程就是将输入的数字图像经处理后与计算机中的所有字进行比较，找出最相近的字就是识别结果。

　　本论文主要介绍了手写体数字识别的一些基本知识和发展概况，然后介绍了基于BP神经网络的手写体数字识别的设计原理，最后本文叙述了利用BP神经网络原理开发的识别数字的系统。此外，在识别之前，预处理中很重要的一个环节。由于原始图像在大小方面存在很大的差异，所以必须进行归一化使得它们具有相同的大小。另外，人们书写的字体往往有不同的倾斜角度和线条粗细，所以要对字体进行书写的倾斜矫正和图像的细化。

关键词:  手写体数字识别；BP神经网络；vc++6.0

Abstract:The digital recognition researches how to treat with and recognize pattern automatically through computer with math arithmetic. Along with the development of computer technology, human need more advanced digital recognition technology. Especially for large numbers of printed data and manuscript, the automatic recognition and input of Chinese characters becomes a stringent task, therefore the digital recognition will have a broad application prospect on literature retrieval, office automation, postal service system, bank bill processing.

　　In order to recognize digital characters, the first task we have to do is feature extraction of a map, after that we have to store the feature in the computer. This process is called "the training". This process compares the hand-written digital’s feature and the stored features in the computer.

This paper describes the recognition of handwritten digits some basic knowledge and develop profiles, and then introduced based on BP neural network recognition of handwritten digital design principles, the last paper describes the use of BP neural network is the development of digital identification systems. In addition, in recognition before the pretreatment is very important aspect. Since the original image in size, there is a big difference, so must be normalized so they have the same size. In addition, people often have different fonts of writing the tilt angle and line thickness, so the font to write the image tilt correction and refinement. 

Keywords: handwritten numeral recognition; BP neural network; vc++6.0