Abstract
CAPTCHAs have become the de-facto standard in providing protection from automated robot attacks against online forms and services. These captchas can take on many forms, combining the use of text, images or other cognitive tasks that are difficult for computers to solve but easy for humans. In this paper, we focus on captchas where the challenge to be solved is an image recognition task. We show that well established image classification techniques and algorithms can be used by attackers to “crack” such captchas. Specifically, we examine three state of the art image recognition captchas, SQ-PIX, ESP-PIX, and ASIRRA, and show that with modern image processing tools, the CAPTCHAs do not provide much security and can be easily circumvented.