Deep Learning Model for Breaking Text Based Captchas
CAPTCHA an abbreviation for Completely Automated Public Turing Test to tell Computers and Humans Apart- invented to stop spammy software from hijacking forums, blog comment sections and also for distinguishing humans from machines. The mostly used common type of captchas is Text captchas, given with several distorted letters so that humans are able to solve them easily within a small amount of time but It is hard for machines to recognize. But with the rapid enhancement of technology in artificial intelligence, hackers are trying to break the security by breaking the captchas and map to their solutions. Moreover, Text based captchas are still being used, assuming that the speed of attack of letters in the captcha are negligible and believed that It won’t affect much to their applications. Hence in our paper, we have implemented a deep learning technique of Convolution neural networks (CNN) for breaking a text based captcha with faster speed and moreover with high accuracy rate. With this demonstration we have proved that text based captchas are no longer provide security and hope for the development of different types of difficulty captchas with the deep learning techniques so that they can’t be automated by hackers and ensures a better security.
Keywords: Text-based captcha, Threats, security, CNN, deep learning algorithm.