An iterative method of palettebased image steganography

1
An iterative method of palette-based image
steganography

• Authors Mei-Yi Wu, Yu-Kun Ho and Jia-Hong Lee
• Source Pattern Recognition Letters, Vol. 25,
  Issue 3, P.301-309, February 2004
• Speaker Shu-Wei Guo
• Date 2004/11/20

2
Outline

• Introduction
• Proposed method
• Experimental results
• Conclusion

3
Introduction

• EZ stego method (Machado, 1997)
• Fridrichs steganographic method Fridrich
  (1999a,b)

4
EZ stego method
Image
0
3
Stegoimage
0
2
Encoding
Palette
Embedded data 00001.
Reorder by luminance
5
Fridrichs steganographic method
Image
Stegoimage
Encoding
Palette
Decoding--- (000) mod 20 (333) mod 21
Embedded data 01001.
6
Fridrichs steganographic method
Encoding
Decoding
7
Fridrichs steganographic method-Example1

• Encoding
• Ci(129,115,28), d1 and the closest
  Cj(128,113,26)
• (129115281)mod 21
• Using Cj to replace Ci such that
  (126113261)mod 20
• Decoding
• (12111326)mod 21------d

8
Fridrichs steganographic method-Example2

• Encoding
• Ci(129,115,28), d0 and the closest
  Cj(128,113,26)
• (129115280)mod 20
• Nothing to do.
• Decoding
• (12911528)mod 20------d

9
Proposed method

• Proposed algorithm
• Definitions
• Cost of removing an entry
• Benefit of creating a new entry
• Improving security

10
Proposed method- Algorithm
Palette
If Benefit gt Cost, then remove the color entry
and add an new color entry.
11
Proposed method- Definitions(1/4)
12
Proposed method- Definitions(2/4)
13
Proposed method- Definitions(3/4)
14
Proposed method- Definitions(4/4)
15
Proposed method- Cost(1/4)

• The embedding error used in Fridrichs method is

16
Proposed method- Cost(2/5)
Palette
17
Proposed method- Cost(3/5)

• The cost derived from the replacement error of
  updating the color entry i to the closest color
  using entry k and the embedding error of the
  updated entry k is

18
Proposed method- Cost(4/5)

• The reference error derived using other entries
  in the palette that select i as their closest
  color during embedding process is

19
Proposed method- Cost(5/5)

• The total error cost is given by

20
Proposed method- Benefit

• Create a new entry Cj which is the closest entry
  to Ci

21
Proposed method- Improving security
22
Experimental results (1-1)
(a) Original image
(b) EZ method
23
Experimental results (1-2)
(a) Original image
(c) Fridrichs method
24
Experimental results (1-3)
(a) Original image
(d) Proposed method
25
Experimental results (2-1)
(a) Original image
(b) EZ method
26
Experimental results (2-2)
(a) Original image
(c) Fridrichs method
27
Experimental results (2-3)
(a) Original image
(d) Proposed method
28
Experimental results (3)
29
Experimental results (4)
Fig. 3. The graphs of the evaluated benefit and
cost functions (a) for the image in Fig. 1(a)
(b) for the image in Fig. 2(a).
30
Experimental results (5)
Fig. 4. The RMS errors with different iterations
by employing the proposed method (a) for the
image in Fig. 1(a) (b) for the image in Fig.
2(a), respectively.
31
Conclusion

• Dynamically and iteratively modifies the palette
  colors to minimize the RMS error
• Decrease the noise generated by Fridrichs method

32
Commend

• The closest color entry may not the best.
• Construct a minimum spanning tree of palette.

33
Commend
0
1
0

34
Proposed method- Algorithm(1/3)
35
Proposed method- Algorithm(2/3)
36
Proposed method- Algorithm(3/3)