Captchacker ProjectAbstractThe Captchacker Project exploits the potential of Support Vector Machines to break visual captchas. We define “simulation-based” and “captcha-based” methods to build our models.
Our method performs extremely well for breaking easily-segmentable captchas, with a robust recognition. We also give a generalization to non easily segmentable – thus harder to break – captchas, using Dynamic Programming. Besides, we explain how to get a large training set without going through the long and boring task of manual labelling.
Finally, our source code is available (for free!) here.
Full paperDownload - BibTeX