In this paper, we concentrate on video watermarking for forensics applications and consider the temporal synchronization problem, which has been overlooked in the literature so far. As a result,
we propose a system that provides temporal synchronization in video
watermarking by using side information at the receiver. Short perceptually-robust representations (also known as robust hash values) of randomly selected frames from the watermarked video regions is derived at the encoder and transmitted to the decoder. Synchronization is then achieved by computing perceptually-representative information of all frames of the received video
at the receiver and finding the "best matching region" via solving
a combinatorial optimization problem efficiently using dynamic programming techniques. A suitably-chosen "robust image hash" function is used to derive the necessary representative information of the video frames; the resulting hash values possess properties of being short in length, computable in real time, and similar (resp. different) for perceptually similar (resp. different) video frames with high probability. We experimentally illustrate the effectiveness of our method against several attacks, which include frame-wise geometric attacks, as well as temporal de-synchronization attacks, such as random temporal interpolation, scene editing, cutting and swapping.
In this paper, we propose a novel semi-blind video watermarking scheme, where we use pseudo-random robust semi-global features of video in the three dimensional wavelet transform domain. We design the watermark sequence via solving an optimization problem, such that the features of the mark-embedded video are the quantized versions of the features of the original video. The exact realizations of the algorithmic parameters are chosen pseudo-randomly
via a secure pseudo-random number generator, whose seed is the secret key, that is known (resp. unknown) by the embedder and the receiver (resp. by the public). We experimentally show the robustness of our algorithm against several attacks, such as conventional signal processing modifications and adversarial estimation attacks.
In this paper, early-stop and Motion Vector (MV) re-use approaches are proposed for the MPEG-2 to H.264 transcoding to reduce the computation of the variable block-size motion estimation. By combining the two approaches, the number of MV search points is reduced by more than 80% without significantly affecting the video quality. The proposed approaches can also be used in fast variable block-size motion estimation for the H.264 video encoding.