1. 1 Robust and transparent watermarking scheme for colour images Speaker : Po-Hung Lai Adviser : Chih-Hung Lin Date : 2009.1.5

2. 2 Article History and Author  Article History Published in IET Image Processing Received on 6th April 2008 Revised on 2nd February 2009  Author K.-C. Liu & C.-H. Chou

3. 3 Outline  Introduction  Estimation of JND profiles of colour images  Watermark embedding and extraction  Simulation results  Conclusions

4. 4 Introduction Business

5. 5 Introduction Business WATERMARK

6. 6 Introduction WA ARK 圖的正確性 Y?N?

7. 7 Introduction  texture, luminance, corner and edge

8. 8 Introduction (visual part)  discrete cosine transform (DCT) by Watson’s perceptual model Li et al. adaptively adjust quantisation step sizes small number of perceptual works can be found  Human Visual System (HVS) watermark was inserted into the perceptually important sub-image

9. 9 Introduction (visual part)  Just noticeable colour difference (JNCD) in the uniform CIELAB colour space perceptually lossless colour quantisation watermark transparency JNCD threshold

10. 10 Introduction (embed part)  Properties of histograms Lin et al. measured the numerous global features of all pixels and three- dimensional feature space  Invisible watermarking method cryptography watermarking

11. 11 Introduction (embed part)  Modify pixels modifying the intensities of pixels in the blue component of the colour image modifying quantisation indices of colour pixels  Spread spectrum watermarking scheme watermark into a four-fork tree in the wavelet domain of the Y component

12. 12 Introduction (defect & improvement)  Fixed step size to quantise Worse robustness  Modification of the quantisation help to improve the performance of the watermarking scheme.

13. 13 Introduction (in this paper)  perceptual redundancy inherent in each wavelet coefficient of the colour image in a more reliable YCbCr

14. 14 Estimation of JND profiles of colour images  Tristimulus values  Perceptually indistinguishable.

15. 15 Estimation of JND profiles of colour images  Color space ， CIELAB (Commission International L’E’ clairage ) RGB ， YUV ， YCbCr to overcome the non-uniform colour metric that had been discussed by MacAdam

16. 16 Estimation of JND profiles of colour images JNCD In CIELAB space any two colours difference is greater than 2.3 Performance of robustness In CIELAB space use JNCD to embed transparent watermark Perceptually lossless colour quantisation Modifying quantisation indices embedded transparent watermark

17. 17 Estimation of JND profiles of colour images  adaptive JNCD (AJNCD) due to local variations in luminance magnitude is considered to estimate the adaptive JNCD (AJNCD) for each colour pixel

18. 18 Estimation of JND profiles of colour images  is a weighting function between AJNCD  is the average background luminance  is the maximum of the four weighted average of luminance

19. 19 Estimation of JND profiles of colour images

20. 20 Estimation of JND profiles of colour images

21. 21 Estimation of JND profiles of colour images  Luminance contrast is given by f1  texture masking effect is given by f2

22. 22 Estimation of JND profiles of colour images

23. 23 Estimation of JND profiles of colour images

24. 24 Estimation of JND profiles of colour images  Choose YCbCr colour space Provides large perceptual redundancy Without degrading the visual quality Can resist JPEG ， JPEG2000 ， H.26x compression attacks

25. 25 Estimation of JND profiles of colour images  is the JND threshold  is the MND threshold  is a lifting factor  minimally noticeable distortion (MND)

26. 26 Estimation of JND profiles of colour images

27. 27 Watermark embedding and extraction

28. 28 Watermark embedding and extraction

29. 29 Watermark embedding and extraction

30. 30 Watermark embedding and extraction

31. 31 Watermark embedding and extraction  is the energy of the estimated JND profile

32. 32 Watermark embedding and extraction  The plots of relative distortion capacity versus band number are obtained

33. 33 Watermark embedding and extraction

34. 34 Watermark embedding and extraction  To guarantee the transparency and robustness  JND tuned by the associated JND value key information facilitate the portability of key

35. 35 Watermark embedding and extraction  full-band JND profiles Bitstream image is too bad JPEG2000 coder reduced to 1 k bytes  Sub-band JND profile tuned colour quantisation JNDs decodedwatermarked coefficients

36. 36 Watermark embedding and extraction  = quantisation index  = quantisation step size

37. 37 Watermark embedding and extraction  = decoded JND threshold

38. 38 Watermark embedding and extraction  ， B=bit sequence  = binary value of 1 or 0  is a random value of 1 or 21 for modulating qZ,i to higher or lower index

39. 39 Watermark embedding and extraction

40. 40 Simulation results

41. 41 Simulation results

42. 42 Simulation results  To verity the robustness

43. 43 Simulation results

44. 44 Simulation results

45. 45 Simulation results

46. 46 Simulation results  Second experiment Huang et al.

47. 47 Simulation results  Stimark software

48. 48 Simulation results

49. 49 Simulation results

50. 50 Simulation results

51. 51 Simulation results

52. 52 Simulation results

53. 53 Conclusions  By using the perceptual redundancy  Watermark transparency is achieved  Quite robust in the face of various attacks.