Reversible data hiding for high quality images using modification of prediction errors

1. Reversible data hiding for high quality images using modification of prediction errors Source: The Journal of Systems and Software, In Press, Corrected Proof, Available online 3 June 2009 Authors: Wien Hong, Tung-Shou Chen, and Chih-Wei Shiu Presenter: Chia-Chun Wu Date: September 4, 2009

2. OUTLINE • INTRODUCTION • RELATED WORKS • PROPOSED SCHEME • EXPERIMENTAL RESULTS • CONCLUSIONS

3. 要解決的問題 • 此篇論文主要是利用相鄰像素值非常相近的特性，以周圍相鄰的像素值來對要進行隱藏的像素值先進行預測的動作，並計算預測值跟實際值的差值，接著結合Ni等人提出來的直方圖無失真資料隱藏的方法，藉由調整預測誤差值來達到達到高容量、低失真的無失真資料隱藏的目的。

4. INTRODUCTION (1/3) Reversible data hiding(Lossless Data Hiding) Cover Image Lossless Embedding Stego-image Secret Data Modification of Prediction Errors (MPE) Lossless Cover Image Lossless Exaction Stego-image Secret Data

5. INTRODUCTION (2/3) • Reversible data hiding (Lossless Data Hiding) • Application: • medical images, military photos, law enforcement • Challenges: • Capacity • Quality

6. INTRODUCTION (3/3) • Reversible data hiding schemes: • Difference expansion • Reversible data embedding using a difference expansion, Jun Tian, IEEE Transactions on Circuits and Systems for Video Technology, vol. 13, no. 8, pp. 890 – 896, Aug. 2003 • Reversible watermark using the difference expansion of a generalized integer transform, Alattar, A.M. IEEE Transactions on Image Processing, vol. 13, no. 8, pp. 1147 - 1156, Aug. 2004 • Adaptive lossless steganographic scheme with centralized difference expansion, C.C. Lee, H.C. Wu, C.S. Tsai, and Y.P. Chu, Pattern Recognition, vol. 41, no. 6, pp. 2097-2106, 2008 • Histogram modification • Reversible data hiding, Z. Ni, Y.Q. Shi, N. Ansari, and W. Su, IEEE Transactions on Circuits and Systems for Video Technology, vol.16, no.3, pp. 354 – 362, March 2006 • Hiding Data Reversibly in an Image via Increasing Differences between Two Neighboring Pixels, C.C. Lin and N.L. Hsueh, IEICE Transactions on Information and Systems, vol. E90–D, no.12, Dec. 2007 • A lossless data hiding scheme based on three-pixel block differences, C.C. Lin and N.L. Hsueh, Pattern Recognition vol. 41, no. 4, pp. 1415 – 1425, April 2008

7. RELATED WORKS (1/3) Ni et al.’s method Original image Histogram ofpixel values Peak point Secret data embedding Zero point 101100 unchanged Extracting Stego image 101100

8. RELATED WORKS (2/3) Thodi and Rodriguez’s method pi = 204 a = 203, b = 205, c = 204, xi = 202 Predicted value xi’ = 2 × pi / 2 xi’ = 2 × 204 / 2 = 204 Prediction error ei between xi and xi’ ei = xi – xi’ ei = xi – xi’= -2 Expanded prediction error Ei = 2 × ei + sj If secret bit sj= 1, Ei = 2 × ei + sj= -3 Stego-pixel yi = xi’ + Ei. ( or yi = xi + ei + sj) yi = 204 + (-3) =201 Embedding phase

9. RELATED WORKS (3/3) Thodi and Rodriguez’s method pi’ = 204 a = 203, b = 205, c = 204, yi = 201 Secret bit sj= LSB(yi), Predicted value yi’ = 2 × pi’ / 2 sj= LSB(yi) = LSB(201) = 1 Expanded prediction error Ei = yi – yi’ yi’ = 2 × 204 / 2 = 204 Ei = 201 – 204 = -3 Prediction error ei = Ei / 2 ei = -3/ 2 = -2 xi = yi’ + ei(or xi = yi – ei – sj). xi = 204 + (-2) = 202 Extracting phase

10. PROPOSE SCHEME (1/6) More suitable Histograms of prediction errors and histogram of pixels in the spatial domain for images Lena and Baboon.

11. PROPOSE SCHEME (2/6) Prediction error ei = xi – pi. Embedding phase

12. PROPOSE SCHEME (3/6) Secret = 1012 154 156 153 156 157 149 148 157 154 158 157 157 158 157 155 Stego image I’ Original image I p2 = 154 c≤ min (a, b) → p1 = 156 e2 = x2 – p2 = -4 : non-embeddable e2 = e2 – 1 = -5 e1 = x1 – p1 = 0 : embeddable e = e + 1 = 1 stopping location L p5 = 150 e5 = x5 – p5 = 7, all secret bits are embedded, set L=(2,2)

13. PROPOSE SCHEME (4/6) Prediction error ei = xi – pi. Extracting phase

14. PROPOSE SCHEME (5/6) 153 156 153 153 154 156 156 154 157 149 150 148 Stego image I’ Original image I c≤ min (a, b) → p1’ = 156 p2’ = 154 e1 = x1’ – p1’ = 1: secret bit = 1 e = e - 1 = 0 e2 = x2’ – p2’ = -5: no secret bit e = e + 1 = -4 p3’ = 149 e3 = x3’ – p3’ = -1: secret bit = 0

15. PROPOSE SCHEME (6/6) 157 156 154 157 149 150 148 154 158 157 157 157 158 157 155 Stego image I’ Original image I p5’ = 150 c≤ min (a, b) → p4’ = 157 e1 = x1’ – p5’ = 7 L =(2,2): all embedded message has been extracted e1 = x1’ – p4’ = 1: secret bit = 1 e = e - 1 = 0

16. EXPERIMENTAL RESULTS (1/6) Experimental results of some commonly used images

17. EXPERIMENTAL RESULTS (2/6) Comparison of PSNR with same embedding capacity

18. EXPERIMENTAL RESULTS (3/6) Experimental results for 23 natural photographic test images sized 768 × 512 (payload is measured in bits).

19. EXPERIMENTAL RESULTS (4/6) Experimental results for test images

20. EXPERIMENTAL RESULTS (5/6) Capacity versus distortion performance of various methods for test images

21. EXPERIMENTAL RESULTS (6/6) Capacity versus distortion performance of various methods for test images

22. CONCLUSIONS • The embedding capacity of proposed scheme is much higher than that of Ni et al.’s method. • The visual quality of the proposed method is better than that of Thodi’s method.

23. 此篇論文之優缺點 • 優點： • 因為一般影像而言，統計完預測誤值的結果後，Peak bin的index都是0，因此，跟Ni.等人的方法比起來，此方法不需額外記錄Zero bin及Peak bin的資訊。 • Ni.等人的方法的方法，不論要藏入的資料量多大，整張影像中每個像素值都會被修改變動到，此方法利用Stopping Location L來記錄Secret Data最後藏完時的座標位址，在這座標之後的像素值就完全不做任何修改或變動，來降低影像失真的程度。 • 缺點： • 跟Ni.等人的方法比起來，此方法要額外記錄及傳送Stopping Location L的資訊給接收端。

24. 研究方向 • 本方法是藉由相鄰的3個像素值來進行預測的動作，若額外多考慮相鄰1個像素值的情況下或是利用其它預測的方法，也許可以提高預測的準確度，當預測的準確度愈高的情況下，Peak bin就愈集中在0的地方，相對的最大可以隱藏的資料量就會提高 (Peak bin的數量決定隱藏量的大小)。