Traitor Tracing After Visible Watermark Removal

Watermarks are often used to protect copyright-protected videos from illegal re-distribution. More specifically, a unique watermark that represents the receiver’s identifier is embedded into the video. In this way, malicious users can be identified when they leak their received version of the video. However, when the watermark is embedded as visible text in the video, it is easy for digital pirates to delete it such that they can no longer be identified. Therefore, copyright owners will benefit from a technique that allows the detection of a visible watermark when it is removed. This paper demonstrates how a visible watermark indirectly generates imperceptible variations over the entire video. As such, these variations in the non-watermarked area can be used as an alternative watermark representation, and thus enable watermark detection even after watermark removal. The experimental results prove that the watermark can be detected as long as the quality of the watermarked video is not significantly reduced, especially if the originally-distributed watermarked video has a high quality. Moreover, the watermark should be embedded into a video with sufficient motion. In conclusion, the proposed technique enables copyright-owners to identify pirates when they illegally distribute visibly-watermarked videos, even when the watermarked area is removed.