DCT fingerprint classifier based group fingerprint

• Published in: 2014 International Conference on Audio, Language and Image Processing pp. 292 - 295
• Main Authors: Zhao Yong, Zhang Aixin, Lu Songnian
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2014
• Subjects: Algorithm design and analysis; capacity of group fingerprint; Classification algorithms; Clustering algorithms; Data security; fingerprint classifier; Fingerprint recognition; group detection; group fingerprint; Multimedia communication
• ISBN: 9781479939022; 1479939021
• DOI: 10.1109/ICALIP.2014.7009803

With the development of business activities, the property rights protection for digital content becomes a hot topic. In prior researches, digital fingerprinting techniques are widely used. They find the illegal distributors by traitor tracing techniques. But in the emerging digital wholesale and retail, it is possible that middlemen collude. Then it becomes a development tendency that the fingerprint is added with the group property which improves the group detection performance. Considering the largely existed COX fingerprint which has low computation cost and easy realization, it will save large resources that we classify them to generate the group fingerprint. We utilize four algorithms (k-means, hierarchical clustering, SOM, FCM) to construct the COX fingerprint classifier through which the group fingerprint generating and group traitor tracing algorithms are implemented. The performance of the group fingerprint on the practicability and security are obtained by the colluding attack and multimedia processing experiments. The experiment results show that it is easy to implement the group fingerprinting schemes by the four algorithms. All the classifiers based group fingerprinting schemes withstand the JPEG compressions. The k-means scheme has superior performance than the other three. Even in the averaging attack experiments, in which the other three got the worst performance, k-means obtains acceptable performance.