Attacking Image Recognition Captchas - A Naive but Effective Approach
Christoph Fritsch, Michael Netter, Andreas Reisser, Günther Pernul · University of Regensburg Publication Server (University of Regensburg) · 2010
Abstract. The landscape of the World Wide Web today consists of a vast amount of services. While most of them are offered for free, the service providers prohibit their malicious usage by automated scripts. To enforce this policy, Captchas have emerged as a reliable method to setup a Turing test to distinguish between human and computers. Image recognition Captchas as one type of Captchas promise high human success rates. In this paper however, we develop an successful approach to attack this type of Captcha. To evaluate our attack we implemented a publicly available tool, which delivers promising results for the Huma-nAuth Captcha and others. Based upon our findings we propose several techniques for improving future versions of image recognition Captchas.