A traceable content-adaptive fingerprinting for multimedia

This paper presents a feasible system of multimedia fingerprinting. A c-TA code which can be constructed flexibly is used. To ensure the applicability of both the maximum collusion size and the maximum number of consumers, the fingerprint length should be considerably increased. We design a content-adaptive watermarking scheme using neural networks to adaptively embed the long-length fingerprint without influencing the imperceptibility significantly. Experimental results show the high detection ratio of traitor tracing.