The use of wavelet watermarking and statistical classification techniques for collusion detection and identification in multimedia forensics
This research proposes a wavelet-based multimedia fingerprint scheme and statistical clustering algorithm for collusion detection and identification. The use of digital multimedia has steadily increased using mediums like the Internet. Encryption is generally used to safeguard content while in transmission, but offers no protection against duplication. Tracing unauthorized content distributors has become an increasing concern for the media industry. Unauthorized duplication, piracy, and illegal redistribution of multimedia content account for several billion dollars in losses every year. It is important to design reliable investigative techniques against unauthorized duplication and propagation, and provide protection in the form of theft deterrence. Some fingerprint embedding schemes are robust against single-user modification attacks. However, a new breed of attacks, known as collusion attacks, have been used to defeat those underlying schemes. These attacks use the combination of multiple fingerprinted copies to create a new version where the underlying fingerprint is highly attenuated, becoming untraceable to the colluders. This research adopts the use of wavelet transforms and statistical classification techniques to effectively identify the set of colluders involved in a collusion attack while maintaining low miss rates and false accusation rates. The experimental results show that the solution is effective in identifying large colluder sets without the knowledge of the number of colluders involved in an attack and the collusion attack used.