SACMark: Spatial-Angle Consistency Watermarking Network for Light Field Image Copyright Protection

Area of Science:

• Computer Vision
• Digital Image Processing
• Machine Learning

Background:

• Light Field (LF) images capture 3D scene information, but their high dimensionality complicates traditional 2D watermarking.
• Existing methods struggle with effective copyright protection for LF images due to data complexity.

Purpose of the Study:

• To introduce a novel deep learning-based watermarking network, SACMark, specifically for Light Field images.
• To enhance copyright protection for LF images by addressing challenges in watermark embedding and extraction.

Main Methods:

• Developed a Spatial-Angular Consistency waterMarking (SACMark) network utilizing spatial-angular feature extraction.
• Implemented consistency matching and fusion strategies for improved feature utilization.
• Employed an encoder-noise-decoder architecture optimized via adversarial training for imperceptibility and robustness.

Main Results:

• SACMark demonstrated high visual quality across various embedding capacities with minimal impact on depth estimation.
• The network showed improved resilience to noise compared to traditional LF watermarking and 2D deep learning methods.
• Essential LF image characteristics were preserved post-watermarking.

Conclusions:

• SACMark offers a promising solution for secure and adaptive LF image protection.
• The proposed method advances the field of digital watermarking for complex multidimensional image data.
• Findings suggest potential for practical applications in safeguarding LF image copyrights.