VisCode: Embedding Information in Visualization Images using Encoder-Decoder Network.

Area of Science:

• Computer Vision
• Information Visualization
• Data Embedding

Background:

• Embedding data within visual representations is crucial for data security and information sharing.
• Existing methods often introduce visual distortions or lack robustness.
• Deep learning offers potential for sophisticated data embedding techniques.

Purpose of the Study:

• To introduce VisCode, a novel framework for imperceptible data embedding into visualization images.
• To develop a robust deep neural network model for encoding and decoding information within visualizations.
• To evaluate the effectiveness and practical applications of VisCode in information visualization.

Main Methods:

• Developed a deep encoder-decoder network trained on visualization images and QR code data.
• Incorporated visualization image saliency features to minimize visual distortion during encoding.
• Designed a saliency-based QR code layout algorithm for efficient large-scale data encoding.

Main Results:

• VisCode successfully embeds data into visualization images with minimal perceptual loss.
• Achieved high decoding success rates across various visualization types.
• Demonstrated robust anti-attack capabilities and efficient time performance.

Conclusions:

• VisCode provides an effective and imperceptible method for data embedding in information visualization.
• The framework is suitable for practical applications requiring secure and integrated data representation.
• Further research can explore advanced embedding techniques and diverse visualization contexts.