Revelio: A Lightweight Captcha Solver Using a Dictionary Based Approach

Abstract

Captcha is an important security measure used by many websites to defend against malicious bot programs. However, with the advancement in the field of computer vision, seemingly complex Captcha schemes have been broken. Although Captcha solving techniques have improved significantly, we observed that many major banking and government websites are still relying on a relatively simple class of text Captchas to counter bot attacks. In this paper, we demonstrate that Captcha schemes deployed on State Bank of India (SBI), Axis bank and Indian Railways (IRCTC) websites can be easily broken using a repertoire of standard image processing techniques. We develop a Captcha solver tool called Revelio which is lightweight, automatic, efficient, and requires minimal labeled data and works in real-time. We evaluate the performance of our tool with the state-of-the-art CNN model on diverse Captcha schemes from 14 major Indian websites. The proposed solver achieves at least 90% accuracy on 10/14 Captcha schemes. Further, we found that for the targeted class of Captcha schemes and a given amount of labeled data, our solver outperforms the CNN based solver.