Abstract: Basic issues on the design of watermarking for authentication are discussed, and a "local authentication" concept based on MPEG motion estimation is proposed. Then, a new twofold authentication scheme based on semantic content is illustrated. Experimental results show that the method is robust to small noise and MPEG recompression, and can detect malicious video semantic content tamper such as face replacing that may be accepted by a common global authentication way. In addition, it can approximately recover the original visual information in the corrupted area to some extent.