扭曲粘连字符验证码识别技术研究

发布时间：2018-10-05 11:33

【摘要】：网络安全问题,比如利用网络机器人批量注册邮箱、论坛灌水、恶意刷票、暴力破解密码等,这样不仅造成资源浪费,影响用户正常地使用,还会给用户的账户安全带来隐患。扭曲粘连的验证码可以增加被机器识别的难度,对于这类验证码的识别研究不仅可以从验证码反识别的角度设计更加安全易用的验证码,还可以及时发现验证码的安全漏洞,提高用户上网安全性。本文对验证码图像预处理技术以及验证码识别理论进行了深入研究。由于扭曲粘连的验证码出格点较多,并且噪声干扰比较严重,基于传统形状上下文等识别方法对于这种类型的验证码形变不变性问题的处理能力很差;针对前一个问题,本文提出了基于相对形状上下文和点模式识别的方法,并对现有的网易邮箱和CSDN的验证码进行识别,有效地解决了网易邮箱和CSDN验证码字符粘连的问题。另外,传统的形状上下文算法还存在着诸如验证码字符轮廓形状和噪声敏感、描述形状特征不准确和对质心的计算方法复杂耗时等缺陷。对此,本文提出了将SIFT(Scale-invariantfeaturetransform)特征和改进的形状上下文相结合的算法,该算法采用SIFT描述子采集验证码形状结构信息,采用部分最小二乘算法消除了错误匹配点。本文以央视网验证码为实验对象,对比分析了实验结果,提高了字符验证码的匹配精度。
[Abstract]:Network security problems, such as using network robots to register email boxes in batches, forum irrigation, malicious ticket brushing, violent password cracking and so on, not only cause waste of resources, affect the normal use of users, but also bring hidden trouble to user's account security. Twisted and conglutinated verification codes can increase the difficulty of being recognized by machines. The research on this kind of verification codes can not only design more secure and easy-to-use verification codes from the point of view of counter-recognition of the verification codes, but also discover the security vulnerabilities of the verification codes in a timely manner. Improve the safety of users online. In this paper, the image preprocessing technology of verification code and the theory of verification code recognition are deeply studied. Because there are many abnormal points in the distorted and conglutinated verification codes, and the noise interference is serious, the traditional recognition methods such as shape context have poor ability to deal with the deformation invariance problem of this type of verification code. In this paper, a method based on relative shape context and point pattern recognition is proposed, and the existing NetEase mailbox and CSDN verification code are recognized, which effectively solves the problem of the character adhesion between NetEase mailbox and CSDN verification code. In addition, the traditional shape context algorithm also has some defects, such as the shape and noise sensitivity of the character, the inaccuracy of describing the shape feature, and the complexity and time consuming of calculating the centroid. In this paper, an algorithm combining SIFT (Scale-invariantfeaturetransform) features with improved shape context is proposed, in which the SIFT descriptor is used to collect the shape structure information of the verification code and the partial least squares algorithm is used to eliminate the error matching points. In this paper, the experimental results are compared and analyzed, and the matching accuracy of character verification codes is improved.
【学位授予单位】：上海应用技术大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP393.08

【参考文献】