Research On Anti-Collusion Quantization Fingerprinting Techniques

Digital fingerprinting is a technique to combat illegal dissemination ofdigital multimedia content. A unique signal, called fingerprint, is embed-ded into every copy before distribution. Once a pirated copy is found, thefingerprint can be extracted to identify the dishonest. Digital fingerprint-ing should resist collusion attacks in which a group of adversaries try toproduce a pirated copy with weakened fingerprints. Existing anti-collusionfingerprinting schemes mainly exploit spread spectrum watermarking tech-nique to embed fingerprints. However, spread spectrum fingerprinting doesnot possess adequate anti-collusion performance in blind detection scenar-ios. Moreover, there is only a little literature dealing with fingerprintingusing quantization watermarking technique which yields competent blinddetection performance.The objective of this thesis is to study quantization fingerprintingtechniques in depth, including the design of schemes and the analysis ofanti-collusion performance and system complexity. In order to facilitatethe research, this thesis ?rst completes the following preliminary stud-ies: First, spread spectrum orthogonal fingerprinting is further studied.A probability density function di?erence factor is used to depict the anti-collusion performance under correlation detection scenarios. The study ofdistance detection scenarios is also included. These results can be used toevaluate the performance of quantization fingerprinting schemes. In addi- tion, quantization operation is reviewed, existing statistical properties ofquantization errors are extended, and methods for generating independentand identically distributed quantization errors under distortion compen-sated subtractive dither quantization are derived. These results providetheoretical foundation for designing quantization fingerprinting schemesand analyzing their anti-collusion performance.The main contributions of this thesis are as follows: First, accordingto statistical properties of quantization error, a quantization orthogonalfingerprinting scheme where fingerprints embedded in di?erent copies aremutually orthogonal is designed. Anti-collusion performance in di?erentcolluder detection scenarios and system complexity are analyzed in detail.A comparison is also made with spread spectrum orthogonal fingerprint-ing. On top of that, quantization binary fingerprinting and quantizationcompound fingerprinting are abstracted from existing quantization finger-printing schemes. Collusion resistance and system complexity of the twoschemes are studied and compared with quantization orthogonal finger-printing. Comparison of these quantization fingerprinting schemes withspread spectrum counterpart is also made. Finally, based on existing re-search, spread transform quantization fingerprinting is further studied. Aspreading parameter and a direction parameter are employed to describethe general implementation of spread transform. Spread transform's e?ec-t on anti-collusion performance of quantization fingerprinting is analyzedwith the help of a spread transform gain factor.