Research And Application On Captcha Recognition And Prevention Based On Neural Network

CAPTCHA is a kind of automated Turing test program, which is used to judge whether the end user is a human being or a computer program. Although there are various kinds of CAPTCHAs, only the text-based CAPTCHA is widely used in commercial websites. The text-based CAPTCHA increases its segmentation resistance with the techniques of the variation of text characters and adding noises, to make sure computer programs are incapable of recognizing the CAPTCHA automatically.The thesis focuses on the main approaches and engineering applications of the recognition and anti-recognition of the text-based CAPTCHA. The main objects of CAPTCHA recognition are the easily segmentable CAPTCHA. Meanwhile, the paper illustrates the application scenarios and approaches of CAPTCHA recognition. After the study on CAPTCHA recognition approaches and its designing process, it summarizes the basic principles and common mistakes in CAPTCHA designing, finally to study the counter recognition techniques and put forward the strong counter recognition strategies.The thesis first analyzes the common properties and design constraints of text-based CAPTCHAs. By reverse engineering an individual CAPTCHA, the paper analyzes the recognition process of an individual CAPTCHA, as well as the procedures and approaches of CAPTCHA recognition. With the preprocessing algorithm, such as graying, thresholding, noise removal and bounding, CAPTCHAs are processed into binary images that consist of text characters only. Then with the vertical projections, the preprocessed CAPTCHAs are split into subimages including only individual text character. Splitting is the key step in the whole process of recognition. Finally the CAPTCHA classification and recognition are finished with the help of pixel counting, vertical projection, horizontal projection, and template correlations. The recognition rate is over 90%. In the state of classification and recognition, neutral network models are established, and the efficiency and accuracy of CAPTCHAs are improved by studying and testing the split CAPTCHAs. After the research and analysis on a great deal of CAPTCHA algorithm and recognition approaches, the paper summarizes the principles of designing strong CAPTCHAs, and further proposes some safe and effective prevention strategies. CAPTCHAs'usability falls down as the CAPTCHA generating algorithms become more and more complicated, and the current widely-used CAPTCHA algorithms are all static. To solve this problem, the thesis put forwards a dynamic CAPTCHA generating strategy, in which by recording the user's performance features the identity of the end-user is determined in accordance with the setting rules and the corresponding CAPTCHAs are generated according to the identity. This strategy confuses the automated recognition steps of robotic programs, and on one hand, improving the usability, and on the other, increasing the security of the CAPTCHAs, which are also the innovation of this paper.