Temporal synchronization in video watermarking

One of the challenges for blind watermark detection is synchronization. Synchronization is the process of identifying the coordinates of an embedded watermark and is crucial in successful watermark detection. If the detector's input is watermarked but synchronization fails, then the embedded watermark will not be detected. In this paper, temporal synchronization for blind video watermark detection is examined by developing new models for watermark embedding and detection. The structure of the watermark, and specifically its key schedule, dramatically affects the ease of synchronization. The new embedder models the construction of the watermark by using a state machine key generator. The key generator can produce time-invariant, time-independent, and time-periodic key schedules as special cases. The watermark detector uses a queue and a state predictor to perform a search to establish and maintain temporal synchronization. These models are general and can be applied to many symmetric blind video watermarking techniques. It is shown that a watermark without temporal redundancy in its key schedule is vulnerable to attacks such as frame dropping and transposition. Using the models, a watermark more resilient against temporal synchronization attacks is designed by adding temporal redundancy in the watermark construction. Experimental results from an implementation of the models are presented.