A Resilient Image Watermarking Using Schur Decomposition, IWT, and Reinforcement Learning Optimization

Abstract

With the ubiquity of digital imagery, there is a heightened concern about verification of ownership and integrity of content. A robust watermark recovery that is tolerant of both intentional and unintentional distortions is still very challenging. In order to overcome these drawbacks, an image watermarking scheme is proposed, which integrates Reinforcement Learning (RL), Integer Wavelet Transform (IWT) and Schur decomposition. The output of this process adaptively controls the embedding process to ensure a desirable compromise between attack resistance and perceptual image quality. The proposed method is based on embedding watermark information in different locations of the Schur U matrix (U3 and U4), where optimization of the watermarking system could be achieved adaptively for both robustness and imperceptibility. The experimental results confirm that the proposed framework has high imperceptibility and strong compression, noise, filtering, blurring, and geometric attacks robustness. Comparative evaluations exhibit better Normalized Correlation (NC) values than other watermarking techniques based on the Schur space. The proposed method differs from the traditional deterministic embedding schemes by using reinforcement learning to adaptively optimize the amount of embedding. The results show high NC value of about 0.9924 under Gaussian noise attack.