Scene tagging: image-based CAPTCHA using image composition and object relationships

In this paper, we propose a new form of image-based CAPTCHA we term "scene tagging". It tests the ability to recognize a relationship between multiple objects in an image that is automatically generated via composition of a background image with multiple irregularly shaped object images, resulting in a large space of possible images and questions without requiring a large object database. This composition process is accompanied by a carefully designed sequence of systematic image distortions that makes it difficult for automated attacks to locate/identify objects present. Automated attacks must recognize all or most objects contained in the image in order to answer a question correctly, thus the proposed approach reduces attack success rates. An experimental study using several widely-used object recognition algorithms (PWD-based template matching, SIFT, SURF) shows that the system is resistant to these attacks with a 2% attack success rate, while a user study shows that the...

• Type: Conference
• Year: 2010
• Where: CCS
• Authors: Peter Matthews, Cliff Changchun Zou
• Keywords: Attack Success Rate, CCS 2010, Image, Large Object Database, Security Privacy