Invisible Watermark based on Self-Supervised Learning

基于自监督学习的不可见水印研究

100,000

Digital media is becoming increasingly common and important in our daily lives and work. Watermark is an effective solution to protect the media from tampering or piracy. Besides traditional methods, which are mainly divided into spatial and transform domains, deep learning-based methods have shown a strong ability to fit different demands in improving watermarking properties, including invisibility, blind detection, and robustness.

This research is dedicated to 1) improving the overall properties of invisible watermarks in specific scenarios via semi-supervised methodology, 2) achieving a trade-off between robustness and invisibility for multi-scenario watermark, and 3) Proposing watermark methods for 3D models with higher performance.