TL;DR
This paper introduces EyePassword, a novel system that uses eye-gaze to enter passwords on an on-screen keyboard, significantly reducing the risk of shoulder-surfing attacks. User studies show it's nearly as fast and accurate as traditional typing, with users preferring it for privacy in public settings.
What Problem Does This Paper Address?
Shoulder-surfing, where an attacker directly observes a user entering sensitive information like passwords or PINs, is a persistent and difficult-to-mitigate security threat. Traditional input methods (keyboards, touchscreens) are vulnerable because the attacker can visually track the user's actions. While existing solutions aim to combat this, they often introduce significant usability issues, requiring users to learn complex new interaction methods or use cumbersome security tokens. This paper tackles the challenge of creating a password entry system that is both secure against shoulder-surfing and maintains the ease of use of traditional password schemes.
Methods
The core innovation is EyePassword, which leverages gaze-tracking technology. Instead of typing on a physical keyboard, users select characters from an on-screen keyboard by simply looking at them. The system uses an eye tracker to determine the user's pupil orientation and map it to a position on the screen.
Key design choices include:
- On-screen keyboard layouts: The paper explores QWERTY, alphabetic, and numeric keypad layouts, noting that the familiar QWERTY layout offers better usability due to learned spatial memory.
- Target size: Keys are designed to be large enough (84 pixels) with sufficient spacing (12 pixels) to minimize false activations due to the inherent uncertainty in eye-tracking accuracy (estimated at 33 pixels).
- Trigger mechanism: Two methods were considered:
- Dwell-based: The user looks at a key for a set duration (450ms in their implementation) to select it. This method is preferred as it doesn't leak timing information.
- Trigger-based: The user looks at a key and then presses a separate trigger key (like the spacebar). This was found to have higher error rates, likely due to the difficulty of coordinating eye gaze and hand movement.
- Feedback: To maintain security, visual feedback (like highlighting the selected key) is minimized. Activation is signaled by a subtle screen flash or an audio beep.
Datasets
The paper does not explicitly mention using pre-existing datasets for training. Instead, it describes a user study where participants entered a set of ten real-world passwords (e.g., "computer," "password," "$dollar$") and a training set of passwords. The passwords were chosen to be representative of common passwords, varying in length (8-9 characters) and including combinations of lowercase, uppercase, numbers, and symbols.
Key Insights and Ideas
- Gaze as a natural input: The paper demonstrates that gaze can be a viable and intuitive input method for sensitive data entry, especially when designed with usability in mind.
- The trade-off between usability and security: Many existing shoulder-surfing countermeasures sacrifice usability. EyePassword aims to strike a better balance by adapting a familiar input paradigm (keyboard entry) to a new modality (gaze).
- QWERTY's enduring advantage: Even in a gaze-based system, the learned spatial layout of the QWERTY keyboard significantly outperforms a purely alphabetic layout, highlighting the power of user familiarity.
- Dwell-based activation is more robust: The dwell-based trigger mechanism proved to be more reliable and less error-prone than a multi-modal trigger approach, likely due to the inherent difficulty in precisely timing eye and hand movements.
- Privacy preference: Users overwhelmingly preferred the gaze-based approach for entering passwords in public settings, indicating a strong desire for enhanced privacy.
Key Results
- Time: Gaze-based password entry took approximately five times longer than traditional keyboard entry (average of 10 seconds vs. ~2500 ms). However, the difference between dwell and trigger-based gaze entry was not statistically significant.
- Accuracy: Error rates for gaze-based entry using the dwell mechanism (both QWERTY and alphabetic layouts) were comparable to traditional keyboard entry (around 3-4%). The trigger-based approach had a significantly higher error rate (15%).
- User Preference: A majority of users (over 80%) preferred the gaze-based approach over traditional keyboard entry when entering passwords in a public place. Users also unanimously preferred the QWERTY layout over the alphabetic layout.
Limitations and Future Work
- Limitations:
- The current implementation is still slower than traditional keyboard entry.
- While effective against typical shoulder-surfing, the system could potentially be defeated by a sophisticated attacker with a camera trained on both the screen and the user's eyes.
- The trigger-based method's high error rate suggests a need for algorithmic improvements to handle eye-hand coordination challenges.
- Future Work:
- Extracting additional entropy from gaze path patterns to further strengthen passwords.
- Developing algorithmic solutions to mitigate the higher error rates observed with the trigger-based mechanism.
- Exploring how to adapt to subtle head movements that might reveal information to an observer.
- Investigating how to make the system secure against advanced attacks that might try to steal eye-tracker data directly.
Summary updated