A Multi-sprite Based Anaglyph 3D Video Watermarking Approach Robust Against Collusion

Abstract

3D video watermarking is far from the maturity of watermarking algorithms dedicated to audio, image or 2D video. In fact, little work has been proposed for anaglyph 3D videos. Such methods have not been robust against malicious attacks such as compression and collusion. The latter one is a very dangerous attack that can be applied on video content and can easily remove an embedded signature to obtain original content. This paper proposes an efficient approach for anaglyph 3D videos which can resist this malicious attack thanks to the use of a static multi-sprite generated from the different sets of frames which compose the original video. First, a sprite is generated for every set of 25 frames. Then every sprite is marked using a hybrid scheme: the discrete wavelet transform based algorithm and the middle significant bit technique, which improve signature invisibility and enhance robustness. Finally, a marked anaglyph 3D video is generated from every marked sprite. The use of multi-sprites as an embedding target provides robustness against collusion attacks and a high quality of marked video reconstruction compared with static mosaics. The experimental results show good robustness against collusion, compression and other attacks such as geometric manipulation and temporal attacks. Besides, the proposed technique presents a high level of invisibility.