EVALUATING A SEGMENTATION-RESISTANT CAPTCHA INSPIRED BY THE HUMAN VISUAL SYSTEM MODEL

Abstract

Visual CAPTCHAs are widely used these days on the Internet as a means of distinguishing between humans and computers. They help protect servers from being flooded by requests from malicious scripts. However, they are not very secure. Numerous image processing algorithms are able to discern the characters used in the CAPTCHAs. It has been suggested that CAPTCHAs can be made more secure if they are distorted in ways that makes segmentation difficult. However, out of all the reviewed distortions present in current CAPTCHAs there are none that allow for a high level of segmentation difficulty. Furthermore, CAPTCHAs also need to be used by humans who may not find certain distortions tolerable. Thus, the problem of selecting a good distortion becomes a tradeoff between user acceptability and computer solvability. It is hypothesized in this paper that rather than use low-level image distortions, optical distortions based on the Gestalt laws of perception that govern human visual system models should be applied. These distortions would ensure widespread user acceptability (as they are based on the internal workings of the HVS), and be very difficult for computers to solve (as HVS perception models have been difficult to implement in computers). This paper aims to explore the feasibility of employing Gestalt-inspired distortion in CAPTCHAs by first implementing a CAPTCHA cracker and then evaluating the performance of some manually generated Gestalt CAPTCHA’s against some existing CAPTCHAs.