1. 1 Watermarking Scheme Capable of Resisting Sensitivity Attack IEEE signal processing letters, vol. 14, no. 2, February. 2007, pp. 125-128 Xinpeng Zhang and Shuozhong Wang

2. 2 Outline 1.Introduction 2.Sensitivity attack 3.Watermark detector 4.Embedding algorithm 5.Detection algorithm 6.Security analysis 7.Experimental results 8.Conclusions

3. 3 Introduction 1.This letter proposes a watermarking scheme capable of defeating the sensitivity attack a novel tailor-made embedding algorithm a corresponding detection mechanism are designed to “mislead” the attackers: get a “fake” signal

4. 4 Sensitivity attack(1/3) 1.Attacker possesses a watermarked image and an available detector 2.Remove or change the embedded without causing serious distortion  Original image Watermark embedding Watermarked image watermark detector Sensitivity attack Not watermarked Watermark

5. 5 Sensitivity attack(2/3) Not watermarked Watermarked Decision boundary Decision boundary : between the “Watermarked” and “Not watermarked” regions is a hyper-plane in a multidimensional space.

6. 6 Sensitivity attack(3/3) : perpendicular to the decision boundary : shows how sensitive the detector is to modification in the direction of each vector : is either +1 or -1 indicates addition or subtraction Subtracts from the watermarked copy with an increasing strength until the detector reports that no watermark is present. Thus, the embedded watermark is removed.

7. 7 Watermark detector Public watermark detector Detection function F > Threshold ≤ Threshold Output 1 (Watermarked) Output 0 (Not watermarked) Black box Test image Watermark detector : provide adequate information about an embedded watermark

8. 8 Embedding algorithm(1/6) C 2,3 C 2,0 C 1,0 C 0,0 C 2,2 C 2,1 C 1,2 C 1,1 C 0,2 C 0,1 DWT 3-levels Watermark IDWT Watermarked image

9. 9 Embedding algorithm(2/6)

10. 10 Embedding algorithm(3/6)

11. 11 Embedding algorithm(4/6) mutually independent and satisfy a standard Gaussian distribution with zero mean and unit standard deviation 1.Generate( )data-groups in a pseudo-random manner 2.The number of elements in each data group is equal to that of the host DWT coefficients, and all elements in the groups :

12. 12 Embedding algorithm(5/6) where All modified DWT coefficients are inversely transformed to yield a watermarked copy : modified DWT coefficients

13. 13 Embedding algorithm(6/6) 1.According to the central limit theorem is very close to its mean 1 All are approximately 0 2.Standard deviation of

14. 14 Detection algorithm (1/2) 1.If output “ Watermarked ” output “ Not watermarked ”

15. 15 Detection algorithm (2/2) 1.If test image contains watermark, all close to 0   2.If test image contains no watermark 3.M  189 avoid excessive distortion

16. 16 Security analysis 1. Attacker does not know and is impossible to estimate by measuring the sensitivities of 2. The detection function of the attacked signal always greater than 0 ( F > 0). This means the sensitivity attack cannot remove the embedded watermark.

17. 17 Experimental results ImageDetection function F Original -1.7  10 4 Watermarked 1.5  10 4 A 960 1280 still image captured by a digital camera was used as the original test signal. The system parameter M=189

18. 18 Conclusions 1.The proposed watermarking scheme is capable of defeating the sensitivity attack. 2.The corresponding detection mechanism can mislead the attackers. 3.The output of detector cannot be used to remove watermark with low distortion.