TL;DR
This paper introduces three novel techniques – Decoy Strokes, Disappearing Strokes, and Line Snaking – to defend graphical passwords against shoulder surfing attacks. Through user studies, they found that Line Snaking offers the best security but Disappearing Strokes provides a better balance of security and usability, making it the recommended choice for general deployment.
What Problem Does This Paper Address?
Graphical passwords, like those used in systems like "Draw-A-Secret" (DAS), are intended to be more memorable than traditional alphanumeric passwords. However, they are particularly vulnerable to "shoulder surfing" attacks, where an attacker can simply observe the user drawing their password over their shoulder. While text passwords are often obscured by asterisks, graphical passwords lack this inherent protection. This paper aims to develop and evaluate practical, software-based defenses against such attacks for recall-based graphical password systems, specifically focusing on those used on portable devices like PDAs, which are often used in public spaces.
Methods
The authors propose three novel shoulder surfing defense techniques for recall-based graphical password systems:
- Decoy Strokes: This technique introduces believable, randomly generated strokes alongside the user's actual password strokes. The goal is to distract the observer's attention away from the real password. The color and thickness of decoy strokes can be user-controlled to help distinguish them from the actual password.
- Disappearing Strokes: This method removes the user's drawn stroke from the screen shortly after it's completed (after the stylus is lifted). This limits the time an attacker has to memorize or capture the password. The duration before disappearance is a tunable parameter.
- Line Snaking: This technique is an extension of Disappearing Strokes, where the stroke begins to disappear while the user is still drawing it, giving the appearance of the line "snaking" towards the stylus. This is intended to provide even less time for an attacker to observe the password. The speed of disappearance is a tunable parameter.
These techniques were implemented on a PDA prototype and evaluated through two user studies. Study 1 focused on the security effectiveness of the techniques against simulated shoulder surfing attacks. Study 2 evaluated the usability (login time, error rates, and user preference) of the more promising techniques.
Datasets
The paper does not introduce new datasets. Instead, it uses pre-defined graphical passwords of varying complexity (weak, medium, strong) based on length and stroke count, as described in previous research [26]. These passwords were used in both user studies for evaluation.
Key Insights and Ideas
- Decoy Strokes are Ineffective: Counter-intuitively, Decoy Strokes performed no better than the undefended system and, in some cases, slightly worse. The authors hypothesize that the relative positions of decoy and real strokes might actually aid attackers in identifying the correct strokes.
- Line Snaking is Secure but Less Usable: Line Snaking proved to be the most effective defense against shoulder surfing in terms of security, significantly reducing the number of stolen strokes and complete passwords. However, it also led to longer login times and more login errors, indicating usability challenges.
- Disappearing Strokes Offer a Good Balance: Disappearing Strokes provided a significant security improvement over the undefended system and Decoy Strokes, while also being perceived as more usable and preferred by participants over Line Snaking.
- Password Strength Matters: Stronger passwords (longer length, more strokes) were inherently harder to shoulder surf, regardless of the defense technique used.
- User Familiarity Mitigates Defense Impact: Participants who were highly familiar with their passwords were less affected by the defense techniques, suggesting that user experience can play a role in the perceived usability of these defenses.
Key Results
User Study 1 (Security): * Line Snaking and Disappearing Strokes significantly reduced the proportion of stolen strokes compared to the control (DAS only) and Decoy Stroke groups. Approximately 40-50% of strokes were stolen with these defenses, versus 77% for the control and Decoy Stroke. * Decoy Strokes offered no statistically significant improvement over the undefended DAS. * Line Snaking and Disappearing Strokes were comparable in their ability to prevent complete password theft. * Password strength was a significant factor, with weak passwords being stolen much more frequently than strong ones.
User Study 2 (Usability): * Login Time: Line Snaking resulted in the highest average login times (e.g., 12.4 seconds for strong passwords), followed by Disappearing Strokes, and then the undefended DAS. * Login Error Rate: Line Snaking also led to more login attempts required for successful logins. * User Preference: A large majority of participants (77%) preferred the Disappearing Stroke technique, citing comfort and confidence. Only 10% preferred Line Snaking, and 13% had no preference.
Limitations and Future Work
Limitations: * The studies used pre-defined passwords, not user-created ones, which might affect the ecological validity of the usability findings. * The "shoulder surfing" scenario was simulated, and participants were not actual attackers with malicious intent. * The defense techniques are primarily effective against casual human observation and may not prevent attacks aided by cameras or recording devices.
Future Work: * Investigate combining defense techniques for potentially enhanced security and usability. * Empirically evaluate these defenses on other graphical password systems like BDAS and Pass-Go. * Explore truly usable zero-knowledge interaction-based techniques. * Further research into improving the Decoy Stroke technique, perhaps by introducing more decoys or modifying their introduction, while carefully avoiding user confusion. * Investigate the impact of user familiarity with passwords on defense effectiveness. * Explore ways to make defenses invulnerable to camera-based attacks, possibly by combining with haptic input devices.
Summary updated