1. 第七章 資訊隱藏
張真誠
國立中正大學資訊工程研究所

2. 影像偽裝術

3. Introduction Images have been widely used in our daily life.
The image security has become an important issue in current computer world.
Image cryptology is a very useful tool to defend the information security.

4. Apply the Traditional Cryptosystem on Images

5. Problems The cipherimage is meaningless. Image Camouflage(影像偽裝)
Image size is huge
Image Compression(影像壓縮)
The decrypted image containing a small distortion is usually acceptable.
Vector Quantization (向量量畫編碼法)

6. Virtual Image Cryptosystem

7. Vector Quantization Encoder

8. Vector Quantization Decoder

9. The Principle of the Virtual Image Cryptosystem
Separate O into a set of vectors {O1, O2, O3,…, Ono}.
Separate V into another set of vectors {V1, V2, V3, … , Vnv}
Let O be the original image
Let {V1, V2, V3, …, Vnv} be the codebook

10. Encryption
Randomly generate the transformed-origin G and the project-direction D.
Project {V1, V2, V3, …, Vnv} to D based on G
Sort the projected results, and obtain {{V’1, V’2, V’3, …, V’nv}

12. Encrypt w, h, no, G, and D into wc, hc, noc, Gc, and Dc by DES-like, respectively.
Encrypt I into Ic, where Ic=IXORX and X is the bit-string containing G, D, G, D,… only.
Hide wc, hc, noc, Gc, Dc, and Ic into the pixels of V.
Cipher Image Vc

13. Decryption

14. Original Image Airplane 512 X 512
Empirical Tests
Test1:
Original Image Airplane X 512

15. Virtual Image
Lena
256 X 256
Cipher Image
Lena
256 X 256
PSNR=37.87dB

16. Decrypted Image Airplane 512 X 512
PSNR=30.22dB

17. Original Image Airplane 512 X 512
Test2:
Original Image Airplane X 512

18. Virtual Image
Lena
360 X 360
Cipher Image
Lena
360 X 360
PSNR=45.13dB

19. Decrypted Image Airplane 512 X 512
PSNR=31.36dB

20. Original Image Peppers 512 X 512
Test3:
Original Image Peppers X 512

21. Virtual Image
Lena
256 X 256
Cipher Image
Lena
256 X 256
PSNR=37.60dB

22. Decrypted Image Peppers 512 X 512
PSNR=29.91dB

23. 資訊隱藏

24. CCC

25. Vector Quantization (VQ) Concept Encoding and Decoding

26. Vector Quantization (VQ)
Image compression technique w h
Index table
Vector Quantization Encoder

27. Vector Quantization (VQ)
Image compression technique w h
Image
Index table
Vector Quantization Decoder

28. 6

29. 標準向量量化編碼法

30. Vector Quantization (VQ) Codebook Training
Codebook generation 1 .
N-1 N
Training Images
Training set
Separating the image to vectors

31. Vector Quantization (VQ) Codebook Training
Codebook generation 1 . 1 .
254
255
N-1 N
Initial codebook
Training set
Codebook initiation

32. Vector Quantization (VQ) Codebook Training1 .
Index sets 1 .
254
255
(1, 2, 5, 9, 45, …)
(101, 179, 201, …)
(8, 27, 38, 19, 200, …)
N-1 N
(23, 0, 67, 198, 224, …)
Codebook Ci
Training set 1 .
Compute mean values
254
255
Replace the old vectors
New Codebook Ci+1
Training using iteration algorithm 10

33. Example
Codebook
To encode an input vector, for example, v = (150,145,121,130)
(1) Compute the distance between v with all vectors in codebook
d(v, cw1) = d(v, cw2) = d(v, cw3) = 112.3
d(v, cw4) = d(v, cw5) = d(v, cw6) = 235.1
d(v, cw7) = d(v, cw8) = 63.2
(2) So, we choose cw8 to replace the input vector v. 11

34. LBG Algorithm
一次訓練 256 個 codewords
做了100次
連續兩次 MSE 之差別已經夠小

35. Full Search (FS) For example: Let the codebook size be 256
# of searched nodes
per image vector = 256
Time consuming
The closest codeword
Image vector
C = {c1, c2, …, cnc}

36. Euclidean Distance The dimensionality of vector = k (= w*h)
An input vector x = (x1, x2, …, xk)
A codeword yi = (yi1, yi2, …, yik)
The Euclidean distance between x and yi

37. 數位浮水印 Watermark ‘National Chung Cheng University’ (64*64 pixels)
Original image (512*512 pixels)

38. 找出相近的Pairs 16

39. 發現 d(CW0, CW8) > TH d(CW13, CW14) > TH CW1 ,CW2 CW4, CW5
藏 1
藏 0
CW1
,CW2
CW4, CW5
CW6, CW7
CW15, CW10
CW12, CW9
Unused
CW0, CW8, CW13, CW14
CW11
,CW3 17

40. Encode Index Table CW0, CW8, CW13, CW14 Unused Index Table
Original Image
Index Table 18

41. Water mark: 1 0 1 0 1 0 0 1 0 1 1 1 1 0 0 1 1 1 1 1 1 1 1 Index Table1 1 1 1 1 1 1
Index Table
Water mark
CW1, CW2,
CW4, CW5
CW6, CW7
CW11, CW3
CW15, CW10
CW12, CW9
藏 1
藏 0 19

42. Water mark: 1 0 1 0 1 0 0 1 0 1 1 1 1 0 0 1 1 1 1 1 1 1 1 Index Table1 1 1 1 1 1 1
Index Table
Water mark
CW1, CW2,
CW4, CW5
CW6, CW7
CW11, CW3
CW15, CW10
CW12, CW9
藏 1
藏 0 20

43. Water mark: 1 0 1 0 1 0 0 1 0 1 1 1 1 0 0 1 1 1 1 1 1 1 1 Index Table1 1 1 1 1 1 1
Index Table
Water mark 21

44. Original image (512*512 pixels)
資訊隱藏 （Data Hiding）
Secret Data:
Hiding
Original image (512*512 pixels)

45. 分 群 CW0 CW1 CW3 CW5 G2 CW2 G1 CW0 CW4 G0 CW1 CW2 CW3 CW4 CW5 Grouping發現
d(CW0, CW4) < TH
d(CW1, CW3, CW5) < TH
CW2
CW5

46. 藏 入 CW1 CW3 CW5 G2 CW2 G1 CW0 CW4 G0 CW2 CW1 CW4 CW2 CW5 CW4
藏 入
Secret bit stream: (101)2= 5
CW1
CW3
CW5 G2
CW2 G1
CW0
CW4 G0
VQ Compression
CW2
CW1
CW0
CW2
CW1
CW4
CW2
CW1
CW4 1 G1 : ︷
CW2 ︸ G0 : ︷
CW0
CW4 ︸ G2 : ︷
CW1
CW3
CW5 ︸
CW2
CW3
CW0 2
CW2
CW3
CW4 3
CW2
CW5
CW0 4
CW2
CW5
CW4 5
After embedding
Codebook
CW2
CW5
CW4

47. Total combinations: 3x3x2x3=54
VQ Compression
CW1
CW3
CW5 G2
CW2 G1
CW0
CW4 G0
CW0
CW3
CW5 G0 : ︷
CW0
CW4 ︸ G2 : ︷
CW1
CW3
CW5 ︸ G2 : ︷
CW1
CW3
CW5 ︸ G2 : ︷
CW1
CW3
CW5 ︸
Secret Bit stream: (101111)2=47
Codebook
CW1
CW3
CW5
CW0
CW4
After embedding
CW4
CW3
CW5
Total combinations: 3x3x2x3=54 25

48. Search-order Coding (SOC)

49. Search-Order Coding (SOC)
Indicator
The compressing steps 31
207
211 8 7 35
P1 =
P2 =
P3 = 0 00 …
P6 = 0 01
Compression codes = …

50. Information hiding on the SOC codes
The proposed scheme:
- Information hiding:
to embed secret data into host image
- Steganography :
to embed secret data into host image and the interceptors will not notice the existence of secret data
- Based on SOC

51. Information hiding on the SOC codes
Main idea:
Ex. receiver receives the compression codes :
SOC
OIV
(original index value)
OIV
SOC
SOC
It means that the embedded secret data is “01100” if SOC is represented to hide “0” and OIV is represented to hide “1”.

52. Information hiding on the SOC codes
Method:
ex. A 3*3 index table: 1 2 3 18 21 31 30 29 32
If the secret data is “ ”, then the hiding position of each bit will be in the raster scan order.

53. Information hiding on the SOC codes
Defined: “0”  embedded into SOC and
“1”  embedded into OIV.
Embedding phase:
SOC ====> there is nothing that needs to change for its
compression codes
hide “0”
SOC ====> translate SOC into OIV
(give up SOC coding and keep the OIV)
hide “1”
OIV ====> there is nothing that needs to change
hide “1”
OIV ====> translate OIV into SOC
ex.
hide “0” 11
(SOC)
+ OIV

54. Information hiding on the SOC codes
compression codes are still OIV:
Ex.
translate SOC into OIV : =>
translate OIV into SOC : =>

55. Information hiding on the SOC codes
Cost table (bits):

56. Information hiding on the SOC codes
Security:
For enhancing the security of our method, the position in the index table for hiding each bit of secret data can be determined by using pseudo random number generator, and the secret data can be encrypted by using traditional cryptography system such as DES or RSA in advance.

57. Experimental results

58. Experimental results

59. Experimental results

60. Experimental results

61. Switching tree coding (STC)

62. Switching-tree coding (STC)
Sheu proposed the STC algorithm in 1999
Re-encode the index table U L
the current index

63. Switching-tree coding (STC)
If P = 7, then P = U
P’ = ‘11’
If P = 10, then P = L
P’ = ‘10’

64. If P = 14, then P = A in index (3)
P’ = ‘01’ || index (3) = ‘ ’
If P = 17, then
P’ = ‘00’ || (17) = ‘ ’
n=5

65. Information Hiding on the STC codes (IHSTC)

66. Information Hiding on the STC codes (IHSTC)
Watermark: …
Index table

67. Information Hiding on the STC codes (IHSTC)
Watermark: …
P’ = ‘00’||(10)
‘00’||(25)
‘00’||(21) …
‘00’||(17)

68. Information Hiding on the STC codes (IHSTC)
Watermark: …
‘10’
P’ = ‘00’||(10)
‘00’||(25)
‘00’||(21) …
‘00’||(17)

69. Information Hiding on the STC codes (IHSTC)
Watermark: …
‘10’
P’ = ‘00’||(10)
‘00’||(25)
‘00’||(21) …
‘00’||(17)
‘10’ ‘00’||(128) …

70. Information Hiding on the STC codes (IHSTC)
Watermark: …
‘11’
P’ = ‘00’||(10)
‘00’||(25)
‘00’||(21) …
‘00’||(17)
‘10’ ‘00’||(128) …
‘10’

71. Three binary connection tree

72. Three binary connection tree
If U-length > L-length
Tree B
If U-length < L-length
Tree C
Otherwise Tree A
Tree B
Tree C

73. Experiment results Image size = 512*512, n = 3 and |H| = 1024
NSTC: 在 index table 中，可藏入之index數
|H|: Secret Information 之長度
Difference: 藏入前與藏入後 bit 數的差異

74. Experiment results Image size = 512*512, n = 3 and |H| = 2048
Image size = 512*512, n = 3 and |H| = NSTC

75. Image size = 512*512, n = 5 and |H| = 1024
Image size = 512*512, n = 5 and |H| = NSTC

76. Conclusions
A novel information-hiding scheme based on a switching-tree coding
The IHSTC system can hide a huge amount of information in the index table
Only a few extra bits are needed to record the corresponding information
The average time needed to hide an information character is seconds
IHSTC -- an efficient and effective scheme for hiding secret information