A Novel Reversible Data Embedding Scheme Using Dissimilar

A Novel Reversible Data Embedding Scheme Using Dissimilar Pairing for Palette-based Images Chair Professor Chin-Chen Chang Feng Chia University Tsing Hua University Chung Cheng University http: //msn. iecs. fcu. edu. tw/~ccc 1

Outline n n Introduction The proposed scheme Experimental results Conclusions 2

Introduction n Information hiding Sender Receiver Compression code: 1000011010… Color palette image Reconstructed image Compression code 1000011010… Secret data: 011 3

Palette image n Color-Palette image Color palette 4

The proposed scheme (1/14) n Luminance-sorted palette Red pixel value Blue pixel value Luminance value Green pixel value 5

The proposed scheme (2/14) n Palette Luminance-sorted palette Sorted palette Luminance value Sort 6

The proposed scheme (3/14) n Dissimilar pairing Sorted palette 2/n n=16 Dissimilar 7

The proposed scheme (4/14) PUL = (RUL, GUL, BUL) PU = (RU, GU, BU) = (6, 9, 7) = (9, 8, 10) = (9, 9, 10) PL= (RL, GL, BL) = (12, 10, 13) The Reference Vector: X：original pixel value predicted pixel value 8

The proposed scheme (5/14) n PLU Predicted indices Cp and Cq Sorted palette Cp Cq 9

The proposed scheme (6/14) n Embedding policies n n n Case-A Case-B Extracting policies n n Case-A' Case-B' 10

The proposed scheme (7/14) n Embedding procedure Cb: index of best cw Case-1: If Cb ≠ Cp and Cb ≠ Cq (Case-A) Sorted palette Db: dissimilar index of Cb and d(Cb, Cp ) < d(Db, Cp) and d(Cb, Cq) < d(Db, Cq) Cp and Cq: indices of P (6, 9, 7) predicted cw U PLU (9, 8, 10) PL (12, 10, 13) Color image Pc=(7, 9, 7) Cb= 10 Dissimilar Db = 2 = (9, 9, 10) d(Cb, Cp) = 29, d(Db, Cp) = 206, d(Cb, Cq) = 73, d(Db, Cq) =218 Cp Dissimilar Dp = 3 =11 Cq =12 Dq = 4 11

The proposed scheme (8/14) n Embedding procedure Original color index Cb= 10 Dissimilar Db = 2 (Case-A) Predicted color index Cp= 11 Cq= 12 Dissimilar Dp = 3 Dq = 4 Embedding policy If S=0 Encode with Cb If S=1 Encode with Db Ex: S=0 Encode with index Cb= 10 Size of codebook: 16, Cb = 10 = (1010)2 Compression codes: 1010 12

The proposed scheme (9/14) n Embedding procedure Cb: best cw Db: dissimilar cw of Cb (Case-B) Case-2: If Cb = Cp or Cb = Cq or d(Cb, Cp ) ≧ d(Db, Cp) or d(Cb, Cq) ≧ d(Db, Cq) Cp and Cq: predicted cw Sorted palette PU (6, 9, 7) PLU (9, 8, 10) PL (12, 10, 13) Color image Pc=(8, 11, 12) Cb= 11 Dissimilar Db = 3 = (9, 9, 10) Cp Dissimilar Dp = 3 =11 Cq =12 Dq = 4 13

The proposed scheme (10/14) n Embedding procedure Predicted color index Original color index Cb= 11 Dissimilar Db = 3 (Case-B) Cp= 11 Cq= 12 Dissimilar Dp = 3 Dq = 4 Embedding policy If S=00 Encode with Cp || Cb If S=01 Encode with Cq || Cb If S=10 Encode with Dp || Cb If S=11 Encode with Dq || Cb Ex: S=01 Encode with indices Cq||Cb= 12 11 Size of codebook: 16, Cq = 12 = (1100)2 Cb=11= (1011)2 Compression codes: 1100 1011 14

The proposed scheme (11/14) n Extracting and restoring policies Compression codes: 1010 (1010)2 = 10 (Case-A') Sorted palette Eb = 10 Dp and Dq: dissimilar cw of Cp and Cq: predicted cw Case-A' : If Eb ≠ Cp and Eb ≠ Cq and Eb ≠ Dp and Eb ≠ Dq 10≠ 11 10≠ 12 10≠ 3 10≠ 4 PU (6, 9, 7) PLU Cp (9, 8, 10) Cq =12 PL (12, 10, 13) Dissimilar Dp = 3 =11 Color image = (9, 9, 10) Dq = 4 15

The proposed scheme (12/14) n Extracting and restoring policies (Case-A') Extracting and restoring policy If d(Eb, Cp) < d(Db, Cp) and d(Eb, Cq) < d(Db, Cq) Get S = 0 and restore the pixel with Eb else Get S=1 and restore the pixel with Db Ex: Get compression index Eb= 10 Dissimilar Db = 2 d(Cb, Cp) = 29, d(Db, Cp) = 206, d(Cb, Cq) = 73, d(Db, Cq) =218 Predicted color index Cp= 11 Cq= 12 Dissimilar Dp = 3 Dq = 4 It satisfies d(Eb, Cp) < d(Db, Cp) and d(Eb, Cq) < d(Db, Cq) Get S = 0 and restore the pixel with Eb 16

The proposed scheme (13/14) n Extracting and restoring policies Compression codes: 1100 1011 (1100)2 = 12 (Case-B') Sorted palette Eb = 12 Dp and Dq: dissimilar cw of Cp and Cq: predicted cw Case-B' : If Eb = Cp or Eb = Cq or Eb = Dp or Eb = Dq Get next compression codes Nb = (1011)2=11 PU (6, 9, 7) PLU Cp (9, 8, 10) Cq =12 PL (12, 10, 13) Dissimilar Dp = 3 =11 Color image = (9, 9, 10) Dq = 4 17

The proposed scheme (14/14) n Extracting and restoring policies (Case-B') Extracting and restoring policy If Eb = Cp If Eb = Cq If Eb = Dp If Eb = Dq S = 00 and restore the pixel with Nb S = 01 and restore the pixel with Nb S = 10 and restore the pixel with Nb S = 11 and restore the pixel with Nb Ex: Get compression indices Eb= 12 Nb= 11 Predicted color index Cp= 11 Cq= 12 Dissimilar Dp = 3 Dq = 4 It satisfies Eb = Cq (12 = 12) Get S = 01 and restore the pixel with Eb 18

Experimental results (1/2) The color palette images with size 512 × 512 Original color palette image PSNR: 35. 451 d. B PSNR: 33. 008 d. B PSNR: 30. 827 d. B Restored color palette image PSNR: 35. 451 d. B PSNR: 33. 008 d. B PSNR: 30. 827 d. B 19

Experimental results (2/2) [1] Chan, C. S. and Chang, C. C. , “A color image hiding scheme based on SMVQ and modulo operator, ” Proceedings of the 13 th International Multi. Media Modelling Conference (MMM 2007), (Cham, T. J. , Cai, J. , Dorai, C. , Rajan, D. , Chua, T. S. and Chia, L. T. Eds. ), Springer-Verlag, Part II, Singapore, 2007, pp. 461– 470. [3] Chang, C. C. , Lin, C. C. and Chen, Y. H. , “Hiding data in color palette images with hybrid strategies, ” to appear in Imaging Science Journal, 2009. [8] Fridrich, J. , “A new steganographic method for palette-based images, ” Proceedings of IS&T PICS Conference, Savannah, Georgia, 1999, pp. 285– 289. [10] Tzeng, C. H. , Yang, Z. F. and Tsai, W. H. , “Adaptive data hiding in palette images by color ordering and mapping with security protection, ” IEEE Transactions on Communications, Vol. 52, No. 5, 2004, pp. 791– 800. 20

Conclusions n n A hybrid method consisting of pixel prediction and dissimilar concepts is conducted in this paper. The proposed scheme has reversibility to restore the stego palette-based image to the original palette-based image. 21

Thank you so much! 22