A reinforcement learning method for decision making process of watermark strength in still images

Scientific Research and Essays

Alimohammad Latif¹ 

关键词:  ;    Digital image watermarking;    reinforcement learning;    watermark strength;    imperceptibility;    robustness.;   

DOI  :  10.5897/SRE10.886

学科分类：社会科学、人文和艺术（综合）

来源: Academic Journals

【 摘 要 】

Digital image watermarking is one of the most important techniques for copyright protection. The robustness and imperceptibility are the basic requirements of digital image watermarking that are contradictory. The key factor that affects both the robustness and imperceptibility is the watermark strength. This paper presents a new method to determine the watermark strength using Reinforcement Learning (RL) in Discrete Cosine Transform (DCT) domain. Thus, finding the watermark strength was formulated as an RL problem. In our study, the defined reinforcement function has two contradictory aspects, the one with positive aspect is with respect to the similarities between the host and watermarked image and the other with negative aspect is with respect to the robustness of the watermark. Therefore, a novel adaptive methodology is introduced to estimate watermark strength to ameliorate both imperceptibility and robustness at the same time. The experimental results show that the proposed RL algorithm for watermark strength estimation improves simultaneously the robustness and imperceptibility of the watermarking scheme.

【 授权许可】

CC BY