基于神经网络的数字验证码识别研究  

Research on recognition of CAPTCHA based on neural network

作　　者：吕刚[1,2] 郝平[1] 

机构地区：[1]浙江工业大学信息工程学院,浙江杭州310032 [2]金华广播电视大学理工学院,浙江金华321000

出　　处：《浙江工业大学学报》2010年第4期433-436,共4页Journal of Zhejiang University of Technology

摘　　要：验证码是网络上普遍采用的一种用于真人交互证明的有效方法.对验证码识别的研究有助于解决硬人工智能问题,促进人工智能领域的进步.现有的研究多是针对一种验证码,通过多种方法进行识别.这类方法对先验知识的依赖很大,识别方法对其他验证码不一定有效,或者需要大量调整来适应新的验证码.为了研究验证码识别算法的适应性问题,通过选取多个具有代表性的网站的验证码图像,基于分割法和Hopfield神经网络进行分析和试验,取得了较好的试验结果.试验结果表明:利用字符图像灰度信息和Hopfield网络可以有效的对可分割的验证码进行分类识别,算法有一定的适应性,并且仅需字符图像的灰度信息既可适应新的验证码,对先验知识的依赖少.CAPTCHA is a commonly adopted effective means for human interactive proofs on Internet. Study in CAPTCHA is helpful to solve hard AI problems. Current methods aim to recognize one type of CAPTACHA via multiple methods. These methods need a mass of prior knowledge and they maybe are not necessarily valid for other CAPTCHA or need massive adjustment to suit new CAPTCHA. In order to study the adaptive problems on recognition algorithm of CAPTCHA, several CAPTCHAs on the representative websites are selected. The recognition algorithm is analyzed and tested based on segmentation method and Hopfield neural network. The test results are good. The experimental results show that the gray information of character image and Hopfield network can be used to recognize the divisible CAPTCHA. This algorithm has a good adaptability. It can adapt the new CAPTCHA with the gray information of character image and it less depends on the prior knowledge.

关 键 词：神经网络 验证码 仿逆规则 文字识别 分割 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]