Template-based 2D and 3D image watermarking techniques using rule and learning approaches

Abstract

The watermark synchronization problem caused by geometric distortion is considered to be the most difficult problem to solve. Recently, new types of distortions are occurring due to various types of content such as 3D content and various content distribution method. New watermarking techniques have been proposed to cope with these new type of distortions. However, the newly proposed watermarking techniques can not avoid synchronization problems caused by geometric distortion. To solve this problem, this dissertation proposes template-based watermarking techniques to protect various types of contents. A rule-based template watermarking method for 2D image which overcomes the problems of invisibility and robustness against geometric distortion of existing 2D image watermarking methods and a rule-based template watermarking for depth-image-based rendering (DIBR) 3D image which complements vulnerabilities of geometric distortion of existing DIBR 3D image watermarking methods are proposed. We also proposed a learning-based template architecture to avoid the disadvantages of existing rule-based methods that require new rules to be designed whenever a new distortion occurs. Experimental results show that the proposed rule-based watermarking techniques outperform the existing techniques and show that the proposed learning-based template can be applied not only to 2D but also to 3D DIBR contents through learning. These results show that learning-based template can quickly cope with copyright infringement in new types of contents.