Research On Survivability Of Hidden Information Using Game Theory

The research on information hiding focuses on mainly two branches:steganography and digital watermarking. Apparently steganography and digitalwatermarking are two different research directions, but the motto of both is the same toconceal the secret message into multimedia. These techniques hope that theconcealment of the secret message does not produce a significant impact on multimedia,and, from this point of view, both techniques are consistent and similar. In thisdissertation, the embedded watermark is equivalent to the secret message. The oppositeof steganography, the steganalysis attack, illegally detects the hidden secret messagefrom multimedia (e.g., the available steganalysis attack can estimate the secret messagehidden using two popular steganographic methods: spread spectrum by Cox, andquantization index modulation by Chen and Wornell) and the attacker can easilydestroys a part of the secret message estimated by using steganalysis attack, with thehoping to achieve the destruction of the secret information but the damage does notaffect the normal use of the attacked (pirated) multimedia. The multimedia protected bythe embedded watermark has some commercial value, the attacker therefore mayemploy steganalysis attack to estimate the embedded watermark in the multimedia, anddestroy a part of the estimated watermark, which will be a threat for the embeddedwatermark and certainly affect the watermark application and its promotion.Aiming at the influence of steganalysis attack on embedded watermark, thisdissertation will apply game theory to model the conflict between the embedment andattack (including unintentional attack and steganalysis attack) of the watermark, and indepth study watermark robustness under unintentional attack and steganalysis attack.Main innovative points of this dissertation are summarized as follows:1. Watermark system model of attack and defense is proposed. Game theory isused to model the conflict between the embedment and attack of the watermark. Theunsupplied application requirement, due to the damage caused by unintentional attackand steganalysis attack to the embedded watermark, is used to establish utility (payoff)function of the hider-attacker game. Nash equilibrium (the optimal building/embedding strategy of the hider, the optimal attack strategy of the attacker) is obtained when thedamage to the embedded watermark satisfies the condition of damage minimum for thehider and maximum for the attacker. Nash equilibrium is the base of the assessment ofwatermark robustness under unintentional attack and steganalysis attack.2. Watermark survivability under unintentional attack and steganalysis attack isgiven. The destructed watermark is inversely proportional to the watermark survivedafter unintentional attack and the attacker destroy the embedded watermark; thereforewatermark survivability is defined and obtained by using Nash equilibrium. Watermarksurvivability is used to assess the damage (caused by unintentional attack andsteganalysis attack) to the embedded watermark and to reflect the ability of theembedded watermark to resist the damage by unintentional attack and steganalysisattack. Experiment results show that watermark survivability can effectively measurethe effect of unintentional attack and steganalysis attack on the embedded watermark.3. Bit error rate (BER) under unintentional attack and steganalysis attack isanalyzed. Watermark survivability and Newton’s generalized binomial theoremcoefficient are used to calculate the average number of the completely destructedwatermark, namely, error correction code can not restore the destructed watermark.BER can also measure the damage caused by unintentional attack and steganalysisattack, but it is difficult to establish the precise relationship between watermark capacity,the unintentional attack destructed watermark, and the number of the unintentionalattack destructed watermark obtained by the attacker. Compared with BER, watermarksurvivability can thereby effectively measure the damage caused by unintentional attackand the attacker.4. Analysis method of watermark detection reliability under unintentional attackand steganalysis attack is proposed. Watermark survivability, the optimal building/embedding strategy of the hider, the optimal attack strategy of the attacker, BER anddetection threshold are used to calculate the probability of false alarm and theprobability of false negative. Experimental results show since both the probability offalse alarm and the probability of false negative only under unintentional attack are lessthan those under unintentional attack and steganalysis attack, watermark detectionreliability under unintentional attack and steganalysis attack is worse than that onlyunder unintentional attack, and that the impact of steganalysis attack on the embedded watermark is not ignored.5. Watermark damage tolerance under unintentional attack and steganalysis attackis analyzed. The optimal embedding strategy of the hider and the applicationrequirement are employed to obtain watermark damage tolerance. Watermarksurvivability, the unintentional attack destructed watermark, the attacker destructedwatermark, watermark damage tolerance are used to derive watermark usefulnesscondition. The embedded watermark meeting the condition can satisfy the applicationrequirement, and also withstand the damage caused by unintentional attack and theattacker using steganalysis attack.