Designing a Face Detection Captcha

DESIGNING A FACE DETECTION CAPTCHA Adam C. Day Completely Automated Tests for Telling Computers and Humans Apart (CAPTCHAs) are quickly becoming a standard for security in every online interface that could be the subject to spam or other exploitation. The majority of today's CAPTCHA technologies rely on text-based images, which present the user with a string of distorted characters and asks the user to type out the characters. The problem with CAPTCHAs is that they are often difficult to solve and can generally be successfully defeated using techniques such as segmentation and optical character recognition. We introduce an image face recognition based CAPTCHA which presents the user with a series of distorted images and the question of deciding which of these images contain a human face. The user is required to click on all presented face images in order to successfully pass the CAPTCHA. The concept relies on the strength of the human ability to detect a face even amongst heavy distortion as well as the inaccuracies and short-comings of face recognition software. The CAPTCHA application was designed with a web interface and deployed on West Virginia University's Computer Science 101 attendance website. To test the success of the CAPTCHA, data for human success rates was compared alongside facial recognition software which attempted to solve the CAPTCHA. The results of the data gathered during testing not only prove the feasibility of face recognition based CAPTCHAs in general, but also provide valuable data regarding human versus computer recognition rates under varying types of image distortion.