VizDefender: Unmasking Visualization Tampering Through Proactive Localization and Intent Inference

Area of Science:

• Computer Science
• Data Science
• Information Security

Background:

• Data visualization integrity is crucial but threatened by sophisticated image editing.
• Tampering techniques include data manipulation and visual encoding manipulation.
• Existing methods lack precise localization and intent analysis.

Purpose of the Study:

• To define and categorize data visualization tampering techniques.
• To present VizDefender, a novel framework for tampering detection and analysis.
• To evaluate the effectiveness of the proposed framework.

Main Methods:

• Developed a semi-fragile watermark module for precise tampered region localization.
• Integrated an intent analysis module using Multimodal Large Language Models (MLLMs).
• Conducted extensive evaluations and user studies.

Main Results:

• The semi-fragile watermark accurately localizes tampered regions while preserving visual quality.
• The MLLM-based module effectively interprets manipulation and infers attacker intent.
• VizDefender demonstrated high effectiveness in detecting and analyzing visualization tampering.

Conclusions:

• VizDefender offers a robust solution for safeguarding data visualization integrity.
• The framework enhances trust in visual data by detecting deceptive manipulations.
• Future work can explore advanced MLLM applications in data security.