TICS: text–image-based semantic CAPTCHA synthesis via multi-condition adversarial learning

Abstract

CAPTCHA is used to distinguish humans from automated programs and plays an important role in multimedia security mechanisms. Traditional CAPTCHA methods like image-based CAPTCHA and text-based CAPTCHA are usually based on word-level understanding, which can be easily cracked due to the recent success of deep learning techniques. To this end, this paper proposes a text–image-based CAPTCHA based on the cognition process and semantic reasoning and a novel model to generate the CAPTCHA. This method synthesizes three features: sentence, object, and location to generate a multi-conditional CAPTCHA that can resist the attack of the classification of CNN. A quantity of experiments has been conducted, and the result showed that the classification of ResNet-50 on the proposed TIC only achieves 3.38% accuracy.