Robust Real Time Breaking of Image CAPTCHAs Using Inception v3 Model

Abstract

Human Interaction Proofs are important to before allowing access to a common publicly available resource. Completely Automated Public Turing Test to Tell Computers and Humans Apart (CAPTCHA) is one of the main technologies to achieve this. CAPTCHA designers and breakers have always been trying to develop methods to outwit each other. Nowadays, a new class of image based CAPTCHAs are appearing on many sites which have been considered difficult to break. Upto 76% accuracy of breaking these CAPTCHAs has been reported in literature. In this paper, use of inception v3 model trained on ImageNet database has been made to break the image based CAPTCHAs. Experiments on a real CAPTCHA dataset of Facebook demonstrate that the proposed technique can break these CAPTCHA in real time with mean accuracy of more than 91%. Introducing inherent flexibility of CAPTCHA recognition can further improve the results.