BAIT: Visual-illusion-inspired Privacy Preservation for Mobile Data Visualization

Paper Title

BAIT: Visual-illusion-inspired Privacy Preservation for Mobile Data Visualization

Publication Info

• Topic area: Privacy-preserving techniques for mobile data visualizations against shoulder surfing attacks.
• Keywords: Privacy-preserving visualization, mobile data visualization, shoulder surfing, visual illusion, human vision system, decoy visualization, spatial frequency, usability, user study.

Background and Problem

• Problem / challenge: Mobile data visualizations are vulnerable to shoulder surfing attacks, where unauthorized observers can quickly extract sensitive information due to the glanceability of visualizations. Existing hardware- and software-based privacy solutions are inadequate, either degrading usability or failing to address the unique challenges of visualizations.
• Significance: Protecting sensitive information displayed on mobile devices is critical, especially in public or semi-public settings, where visualizations often reveal personal or professional data.
• Motivation and related work: Prior methods, such as privacy films and masking schemes, have limitations like reduced readability, manual configurations, and insufficient protection for highly glanceable visualizations. This paper builds on insights from human vision and visual illusions to address these gaps.

Solution

• Proposed approach: BAIT (Better privacy preservAtion for mobIle data visualizaTion), a fully automated method that overlays decoy visualizations on original visualizations to mislead shoulder surfers while maintaining readability for legitimate users.
• Novelty:
  1. Introduction of a decoy visualization framework leveraging visual illusions and human vision system (HVS) characteristics.
  2. Optimization of privacy-preserving visualizations using a perception-driven model balancing visibility for users and privacy against attackers.
  3. Empirical validation through user studies demonstrating effectiveness and usability across multiple visualization types.
• Procedure and key techniques:
  1. Visual variable identification: Define visualization-dependent (shape, position, tilt, size) and visualization-agnostic (color, spatial frequency) channels for decoy design.
  2. Decoy generation: Randomize visualization-dependent channels and adjust visualization-agnostic channels (e.g., color, spatial frequency) based on HVS principles.
  3. Perception-driven optimization: Use image similarity metrics (VSI, MS-SSIM) to maximize user visibility at close distances and decoy effectiveness at far distances.
  4. Implementation: Generate privacy-preserving visualizations for bar charts, line charts, scatter plots, and pie charts.

Results

• Concrete findings:
  • BAIT achieved user recognition accuracy of 96% at close distances (30 cm) and reduced shoulder surfer recognition accuracy to 4% at far distances (90 cm).
  • Performance was consistent across bar, line, scatter, and pie charts.
  • NASA-TLX workload score averaged 27.22/100, indicating low cognitive demand.
• Advantage over baselines:
  • BAIT significantly outperformed unprocessed visualizations (UV) and masking schemes (MS), reducing shoulder surfer accuracy to near-zero while maintaining high readability for users.
  • Unlike MS, BAIT actively misled attackers rather than merely obscuring content.
• Experiments / evaluation:
  • Study 1 (Controlled Lab): 32 participants evaluated BAIT against baselines across four visualization types at two distances (30 cm, 90 cm). Recognition accuracy and statistical tests demonstrated BAIT’s effectiveness.
  • Study 2 (Real-world Usability): 12 participants tested BAIT in real-world scenarios (e.g., cafés, corridors). Usability was assessed using PSSUQ and NASA-TLX, with qualitative feedback highlighting ease of use and practical value.
• Limitations and future work:
  • Limited to four visualization types; future work will explore applicability to other types (e.g., graphs, infographics) and animated visualizations.
  • Small sample size (32 in Study 1, 12 in Study 2) limits statistical generalizability.
  • Future research could explore other visual illusions (e.g., crowding effect) and extend evaluations to diverse user populations and devices.

Summary

This paper introduces BAIT, a novel method for privacy-preserving mobile data visualizations that overlays decoy visualizations on original charts to mislead shoulder surfers while maintaining readability for legitimate users. By leveraging visual illusions and human vision system characteristics, BAIT achieves a balance between privacy and usability, as validated through controlled experiments and real-world usability studies. The method outperforms existing solutions and offers a practical, software-based alternative to hardware privacy screens, making it a promising approach for protecting sensitive visualizations in public settings. Future work will explore broader visualization types, animated visualizations, and larger-scale evaluations.