Embedding and detecting a watermark in an audio or video bit stream

摘要

In watermarking digital audio/video data, typically a small item of identification information is embedded into the signal. In various watermarking techniques, the watermark data are arranged in a cyclical manner in the audio data. However, when repetitive identification information is watermarked into the signal the difference between the unwatermarked audio and the watermarked audio exhibits periodic patterns which may become audible. To avoid increased audibility of such watermarks, the identification data are scrambled using an n-bit-to-m-bit mapping, where m>n, and from the candidate m-bit words one is selected randomly.