Machine learning attacks against the ASIRRA CAPTCHA

The ASIRRA CAPTCHA [EDHS2007], recently proposed at ACM CCS 2007, relies on the problem of distinguishing images of cats and dogs (a task that humans are very good at). The security of ASIRRA is based on the presumed difficulty of classifying these images automatically. In this paper, we describe a classifier which is 80.6% accurate in telling apart the images of cats and dogs used in ASIRRA. This classifier is a support-vector machine classifier trained on color and texture features extracted from images. Our classifier allows us to solve a 12-image ASIRRA challenge automatically with probability 7.5%. This probability of success is significantly higher than the estimate given in [EDHS2007] for machine vision attacks. Our results should inform the choice of security parameters in future deployments of ASIRRA.

Citation

Golle, P. Machine learning attacks against the ASIRRA CAPTCHA. 15th Annual ACM Conference on Computer and Communications Security (CCS 2008); 2008 October 27-31; Alexandria, VA. NY: ACM; 2008; 535-542.