Watermarking scar as an ultimate copy protection

Strong and powerful attacks are able to make the watermark extraction impossible even if there is some watermarking. In this work, we call this residual information the scar and we use it to prove the existence of an attacked watermark. We evaluate the Scar by the mutual information between the embedded watermark and the attacked copies. It is well known that the mutual information between two random variables gives an evaluation of the shared information between the two variables. Thus, the attack can compromise the correct decoding by removing partialy the watermarking information. However, it is possible to make the watermarking existence proof possible when the mutual information between the embedded message and the attacked copies is above a certain threshold. We propose a practical way to use the scar by measuring the correlation between the attacked watermark and the original one.