Research Cracking Technology Text Verification Code Characters

As the result of the booming Internet, CAPTCHA(Completely Automated Public Turing test to tell Computers and Humans Apart) is used to protect websites from malicious program’s attacking. The study on cracking CAPTCHA can not only help finding CAPTCHA design flaws and improving the security of websites, but also accelerate the development of artificial intelligence, handwritten character recognition, license plate recognition and the like.The categories of currently used CAPTCHA are listed and relevant researches are introduced. According to the fundamental cracking steps, algorithms are studied in this dissertation. The main work is as follows:(1) CAPTCHA image preprocessing algorithms are studied. Firstly, a denoising algorithm based on connected-component is proposed, which is not only harmless to the character information, but also removes the big block noise successfully. Secondly, two algorithms are proposed to segment the touching characters. One is based on the width of the character, while the other is on the basis of the character numbers.(2) The basic theory of artificial neural network is briefly introduced. To speed up the calculation, an appropriate sigmoid function is selected. Based on the convolutional neural network, a character recognition algorithm is studied to implement the simplified convolutional neural network structure.(3) An improved character recognition algorithm based convolutional neural network is proposed and implemented in C++programming language. This improved algorithm is tested by UNIPEN character library. The experimental results indicate that the character can be well recognized with faster convergence speed and higher precision.(4) A software tool is developed for the convenience of debugging designed algorithms. Solutions for CAPTCHAs with three different background noise, such as CCB, China Unicom and douban’s CAPTCHAs, are made. The crack rats are88.67%,93.5%and23%.