Visual CAPTCHA with handwritten image analysis

Abstract

By convention, CAPTCHA is an automated test that humans can pass but current computer programs can't. In general, the research on CAPTCHA and Human Interactive Proofs is focusing on those recognition tasks that are harder for machines than for humans. The recognition of unconstrained handwriting continues to be a difficult task for computers and handwritten image analysis is still an unsolved problem. Therefore, handwriting recognition provides a reasonable gap between humans and machines that could be exploited and used for new CAPTCHA challenges. In this paper we use handwritten word images and explore Gestalt psychology to motivate our image transformations. The deformation methods are individually described and results are presented and compared to other traditional handwritten image transformations. Several applications for Web services would find our handwritten CAPTCHA an excellent biometric for online security and a way of defending online services against abusive attacks. © Springer-Verlag Berlin Heidelberg 2005.