Watermarking for large-scale video distribution through out-of-the-loop frame replacement

Forensic watermarking is used in large-scale video-distribution applications to track digital pirates after they illegally leak videos. Existing methods either have a high embedding complexity or have troubles providing imperceptibility and robustness. Therefore, this paper proposes a novel watermarking method that has a low embedding complexity, as well as sufficient imperceptibility and robustness. The main novelty is that frames are replaced out of the loop, which has a very low complexity. This is inspired by A/B watermarking, yet applied on frame level rather than video-segment level. The resulting drift-error artifacts are used for robust watermark detection. Additionally, a faster watermark detection method is proposed at the cost of a decrease in embedding capacity. We show that the drift errors caused by frame replacement are imperceptible, and that our method has a negligible impact on the video bitrate. Moreover, we demonstrate a high level of robustness. These interesting properties come at the cost of a non-blind detection with a relatively high complexity, that is partially solved by the proposed fast detection method. Additionally, a high detection complexity is not necessarily a problem since detection occurs infrequently and recent speed-up solutions exist. In conclusion, the proposed method is a novel and scalable solution that enables secure large-scale video distribution