Balancing security and usability in a video CAPTCHA
- When?
- Wednesday 19 November 2008, 14:00 to 15:00
- Where?
- 39BB02
- Open to:
- Staff, Students
Dr Richard Zanibbi, Rochester Institute of Technology, NY, USA
Abstract:
CAPTCHAs' are online tests used to prevent programs from automatically acquiring resources intended for live persons. For example, some programs collect email accounts on the internet for the purpose of sending spam emails. To avoid this, many web-based email providers require noisy text images to be transcribed before an email account may be set up or accessed, under the assumption that people are significantly better at this task than existing pattern recognition algorithms. We present a first attempt at using content-based video tagging as a CAPTCHA task. In our task, the user must submit three tags describing the content of a video, one of which must belong to a set of 'correct' tags for the challenge to be passed. Correct responses are defined using tags associated with videos in a public database (YouTube.com). In a user study involving 184 participants, we were able to increase the mean human success rate on our Video CAPTCHA from roughly 70% to 90%, while keeping the mean success rate of an attack using tags estimated to be most frequent fixed at around 13%. Through different parameterizations of the challenge generation and tag matching algorithms, we were able to reduce the success rate of the attack to 2%, while still increasing the human success rate from 70% to 75%. Our initial results indicate that the usability and security of our video CAPTCHA is comparable to that of existing CAPTCHAs.
Notes:
Dr Richard Zanibbi is an Assistant Professor in the Department of Computer Science at the Rochester Institute of Technology (USA), where he directs the Document and Pattern Recognition Lab (http://www.cs.rit.edu/~rlaz/dprl.html). His research interests include pattern recognition, machine learning, document recognition, and human-computer interaction. Among other projects, Richard contributed to the open source pen-based math editing prototype FFES/DRACULAE.Richard Zanibbi is an Assistant Professor in the Department of Computer Science at the Rochester Institute of Technology (USA), where he directs the Document and Pattern Recognition Lab (http://www.cs.rit.edu/~rlaz/dprl.html). His research interests include pattern recognition, machine learning, document recognition, and human-computer interaction. Among other projects, Richard contributed to the open source pen-based math editing prototype FFES/DRACULAE.
