Recognition Of Distorted And Merged Text-based CAPTCHA

As the importance of network increases in life, network security is gaining attention from all sectors, CAPTCHA is one of the effective means to maintain network security. CAPTCHA as a Turing test, its purpose is to distinguish the user is a human or computer, CAPTCHA can effectively block attacks and maintain network security. CAPTCHA recognition can discover the security vulnerabilities in time and promote the design level of CAPTCHA, which has great significance.Currently almost all CAPTCHA recognition methods are based on segmentation, such technologies are quite mature, but not suitable for distorted and merged CAPTCHA, the research of distorted and merged CAPTCHA recognition is still in the relatively empty stage. To solve this problem, this paper proposes methods and uses real data to verify the effectiveness.The main research work and contributions are as follows:l.We conduct the theoretical and practical research on the key technologies of traditional CAPTCHA recognition, such as pre-processing technologies, character segmentation technologies, feature extraction technologies, machine learning technologies and shape-context matching technologies.2.We present a recognition method for distorted and merged CPATCHA. We design and extract DENSE SIFT feature, and then utilize the feature matching to obtain matching point set. Matching indexes are designed in advance, lastly we get the final result through queue-analysis algorithm. Our method can effectively recognize the distorted and merged CAPTCHA.3.We present a recognition method based on Markov Random Field for distorted and merged CAPTCHA. First, we build the MRF model in CAPTCHA spatial domain, and then solve the MRF by the improved belief propagation algorithm. Matching indexes are designed in advance, lastly the algorithm recognizes characters one by one and erases the corresponding structure to acquire the final result. Our method can effectively recognize the distorted and merged CAPTCHA.The recognition methods for distorted and merged CAPTCHA in this paper are tested on CAPTCHAs of real websites and have a good performance.