A diagram showing a simple dynamic watermarking system. The watermark W, the input program P and a key K are fed into the embedder. The key, in this case, is a secret input to the program. This input causes a specific path through the program execution to be taken. The embedder outputs a new program P' which, when executed with the secret input, generates the watermark. A recogniser inspects the program P' while running, looking for value of the watermark variable and outputs the original watermark W.
In comparison to a static system, a dynamic watermarking system needs to execute the program in order to retrieve the watermark. The watermark is stored in the semantics of the program rather than the syntax.
Dynamic watermarks should, in theory, be resilient to semantics-preserving transformations.
