图像验证码识别技术的研究与实现.docVIP

图像验证码识别技术的研究与实现 学校代码 10126 学号分 类 号 密级 本科毕业论文(设计) 图像验证码识别技术的研究与实现 学院、系 计算机学院 专业名称 计算机科学与技术 年 级 2010级 学生姓名 谷红岩 指导教师 魏宏喜 2014 年 月 日 内蒙古大学本科毕业论文(设计) 第 1 页 图像验证码识别技术的研究与实现 摘要:验证码是全自动区分计算机和人类的图灵测试，在维护网络安全，防止机器自动攻击方面做出了巨大贡献，但部分验证码仍可以被现有的模式识别技术破解。本文提出了一种基于BP神经网络的验证码识别方法，能对数字验证码和兼有数字、英文字母的验证码图像进行高效、快速的识别。首先，要对图像进行预处理，比如灰度化、二值化、去噪等;其次，采用连通域分割法获得单个字符图像，再归一化;再次，根据投影特性和轮廓特征，对归一化后的字符进行特征提取，构建并训练BP神经网络;最后，采用交叉验证的方式，实现字符识别。本文基于MATLAB环境来完成此项研究。实验结果表明，对数字验证码的识别正确率达95%以上，而对兼有数字、英文字母的验证码识别正确率达79%;在速度和准确率上都取得了理想的效果。 关键词:验证码，BP神经网络，交叉验证，MATLAB 内蒙古大学本科毕业论文(设计) 第 1 页 The research and implementation on CAPTCHA recognition technology Author:GU Hongyan Tutors:WEI Hongxi Abstract: CAPTCHA is a fully Turing test of automatic distinguish between computers and humans.It has made great contributions on maintaining the Internet security and preventing malicious machine attack. However, some CAPTCHAs can also be recognized by the existing pattern recognition techniques. This paper based on image recognition and neural networks, an effective and rapid CAPTCHA recognition way is put forward which has numbers and English letters. First of all, pre-operating is the first step to achieve this function, including graying, binarization and removing noise. Secondly, images with single character are obtained by a segmentation algorithm combining connected domain, and then normalize the characters. Thirdly, features of characters are extracted based on projection and outline features, the BP neural networks are built and trained. Finally, in the character recognition phase, the trained NNs are tested with the method of cross-validation. Based on the MATLAB platform, a large number of CAPTCHA images from different sites are processed. The experiments show that our attack to number letters achieves a success rate of 95%, Besides, to both numbers and English letters achieves a success rate of 79%.The speed and the success rate of the results are desirable. Key words: CAPTCHA , BP neur