Traitor Tracing for Image Fingerprinting

• Main Authors: Yu-Tzu Lin, 林育慈
• Other Authors: 吳家麟
• Format: Others
• Language: en_US
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/73052383610834723396

博士 === 國立臺灣大學 === 資訊工程學研究所 === 95 === The demand of digital data protection is getting stronger dramatically nowadays as more and more digital contents are used on computers and distributed via the Internet. Digital fingerprinting is a technology designed to help protect digital assets and identify security leak sources. This dissertation addresses the problem of traitor tracing for image fingerprinting, including two major issues: watermarking strategies for image fingerprinting and the design of collusion-secure fingerprinting codes. In the fingerprint-watermarking, we propose a content-adaptive watermarking strategy, which can adaptively decide watermarking strengths and select suitable embedding positions according to the content based on a neural network and an optimization procedure respectively. It maintains the equilibrium of the robustness and the imperceptibility when embedding long fingerprints without the effort to deal with human visual models. Besides, the collusion-resistance is considered to resist collusion attacks. Experimental results show the feasibility of the proposed watermarking algorithm. In the fingerprint-designing, we derive a code construction scheme which constructs fingerprint in a concatenated way based on a user grouping policy. The concatenated construction method is analyzed and proven to be efficient in collusion-secure traceability codes: the larger alphabet size of the outer code guarantees the higher tracing ability and the binary inner code makes the fingerprint satisfying the “Strict Marking Assumption”. The group-based fingerprinting scheme can further improve the tracing ability under the concatenated code construction. Based on these two ideas of the fingerprint designing, we propose three collusion-secure traceability codes: the group-oriented traceability code, the hierarchical traceability code, and the member-exclusive traceability code. All of them successfully disperse the decoding efforts over the outer code and the inner code by the grouping structure. Both theoretical analyses and the practical implementation show good performances of error rates and collusion-resiliency. To sum up, the watermarking and code construction schemes developed in this dissertation provide more flexibility and higher collusion-resiliency than existing solutions, and thus offer a better match for the challenges of the traitor tracing for image fingerprinting.