Image Steganography Method Based on PVD and Modulus Function
Feng Pan 1),2), Jun Li1),Xiaoyuan Yang2)3)
1)
(Key Laboratory of Network & Information Security under the Chinese Armed Police Force,Electronic Department,
Engineering College of the Armed Police Force, Xi’an, China)
2)
(Key Laboratory of Network & Information Security of the Ministry of Education, Xidian University, Xi’an, China)
3)
(Institution of Network & Information Security under Engineering College of the Armed Police Force, Xi’an, China)
E-mail: lijun9250lj@163.com
Abstract—The traditional steganography methods with Pixel
Value Differencing (PVD) usually only deal with the PVD in
the horizontal direction but ignore the vertical one. In this
paper, the cover image is divided into 2×2 nonoverlapping
blocks, then the modulus function method is used in horizontal
direction and the simple PVD method is used in vertical
direction for steganography. In addition, the falling-off-
boundary problem is solved. Experimental results show that
the proposed approach provides higher stego-image quality,
larger embedding capacity, and can against common attacks.
Keywords- Pixel Value Differencing(PVD); steganography;
modulus function;Embedding capacity
I. INTRODUCTION
In recent years, with the great development of Internet,
information security issues become more and more
important. Steganography technology as an emerging
technology has become an important field of information
security Components. Currently, there are many image
steganography method such as: paper[1] proposed a method
based on PVD (Pixel Value Differencing), which using the
difference value of two adjacent pixels of cover image to
determine the number of bits that could embed into the
cover image, unfortunately, this method has low embedding
capacity, poor stego image quality and also low security;
paper [2] proposed a kind of program based on modulus
function, which can significantly improve the quality of
stego image and resistant to the attacks of literature [3], but
the embedding capacity is limited; paper [4] proposed a
method that embedding secret message both in the
horizontal and vertical direction, which is also has poor
performance in embedding capacity, security and the stego
image quality because it embedding message apply simple
PVD method[1]. Make use of good features in literature [1,
2, 4], this paper presents a steganography method based on
pixel value difference and modulus function.
II. RELATIVE ALGORITHMS
A. Original method of PVD
In PVD scheme, the cover image is divided into non-
overlapping blocks, where each block consists of two
consecutive pixels. The data hiding procedure is
independent in each block. Denote the pixel pair in some
block as Pi and Pjˈthe difference d is calculated as d=Pj−Pi
978-1-4577-0321-8/11/$26.00 ©2011 IEEE
282
ˈwith |d| ę [0, 255]. This range is divided into r sub-
regions Rk (k=1,2,Ă,r), the width of each sub-region are
power of 2. The number of bits embedding in the region of
Rk is n = log 2 ( wk ) , where wk is the width of the k-th sub-
region.
B. A method of PVD based on modulus function
Like the same as literature [1], in paper [2], the cover
image is divided into non-overlapping blocks, b is the
decimal value of secret information, calculate:
Frem = ( pi + p j ) mod(2n ) (1)
Specific embedding algorithm was described in paper
[2]. When extract the secret information, first of all ,
calculate the number of bits each pixel can be embedded
n = log 2 ( wk ) , then using equation (1) to calculate the decimal
value of the embedded information, and finally change the
decimal value Frem to n bits secret message.
III. STEGANOGRAPHY METHOD BASED ON PVD AND
MODULUS FUNCTION
The basic processing unit of this paper’s method is 2 × 2
pixel blocks, two pairs of horizontal direction’s pixels and a
pair of vertical direction’s pixels are using the method in
paper [2] and [1] to embed message respectively. In order to
extract the secret information correctly, a modification to
another pair of vertical direction’s pixels is needed; if there
has falling-oơ-boundary problems, the value of pixels will
have to be adjusted; if the above operation doesn’t work, this
block of pixels is not used to embed in the vertical direction.
A. Embedding process
Embedding process steps are as follows:
(1). Let the original cover image from left to right, from
top to bottom into non-overlapping blocks with 2 × 2
pixels. For an image with M×N (length and width), we can
get nblocks =  M / 2  ×  N / 2  blocks. The position and value of
4 pixels in a block are as figure 1:
A B
C D
Figure 1. A example of 2 × 2 block
(2). Calculate d1= |B−A| , which belongs to the region of
Rd1, the number of embedding bits is n1; calculate
 d2=|D−C| , which belongs to the region of Rd2, the number Then this image had taken two rows to embed information
of embedding bits is n2. Using the modulus function’s about vertical-not-available blocks (i.e., x=1), the total
method of paper [2] to embed message for AB and CD, number of vertical-not-available blocks is 1, the label is
while the pixel value after embedding are A1, B1, C1 and D1 1288.
respectively. (2). Let A2 B2C2 D2 denotes a secret message block.
(3). Calculate d3=C1-A1, which belongs to a region of Extract secret bits of horizontal direction’s 2 pairs A2B2 and
Rd3, the number of bits can be embedded is n3; calculate C2D2 with the method of [2]. If this block is not a vertical-
d4=D1-B1, which belongs to a region of Rd4, the number of not-available block, then calculate d3=C2-A2 ˈ d4=D2-B2 ˈ
bits can be embedded is n4. If n3İn4, select A1C1 for the the corresponding embedding capacities are n3 and n4. When
vertical direction’s pixels to embed message with the n3 İ n4, it means that A2C2 had embedded message,
method of paper [1], otherwise B1D1 is the choice. In order otherwise the message was embedded in B2D2, so we can
to extract the secret message correctly, a modification to extract message with the method shown in [1]. If the block is
another pair of vertical direction’s pixels is needed. Let’s a vertical-not-available block, there has no secret message to
take A1C1 for example. Suppose that A2C2 is the value of extract in the vertical direction.
A1C1 that has been embedded message, then we should
change B1D1 to B2D2 as follows: IV. EXPERIMENTS AND ANALYSIS
B2 = B1 − ( A2 − A1 ) (2) This section is the simulation of the proposed method,
D2 = D1 − (C2 − C1 ) (3) the following data are the average of several experiments
If B2D2 has not falling-oơ-boundary problems, then the with 512 × 512 standard gray-scale images. We will suing
embedding process is completed, otherwise the process go to embedding capacity, stego image’s visual quality[6], and
step (4). program’s security to scale our steganography’s
(4). If B2D2 has falling-oơ-boundary , select performance. The comparison of paper [1, 2] with the
n = max( n1 , n2 ) , we revise pixels as follows: proposed method in embedding capacity and PSNR (Peak
1) If B2 < 0 || D2 < 0 ,then modify A2 = A2 + 2n / 2 , Signal Noise Ratio) are shown in Table 1 and 2.
B2 = B2 + 2n / 2 , C2 = C2 + 2n / 2 , D2 = D2 + 2n / 2 , if A2 B 2 C 2 D 2 TABLE I. COMPARISON OF PAPER [1] WITH THE PROPOSED METHOD
still falling-off-boundary, go to (3); Method in [1] Proposed method
2) If B2 > 255 || D2 > 255 ,then modify A2 = A2 − 2n / 2 , Cover vertical-
image Capacity Capacity/ not-
PSNR/dB PSNR/dB
B2 = B2 − 2n / 2 , C2 = C2 − 2n / 2 , D2 = D2 − 2n / 2 , if A2 B 2 C 2 D 2 /bits bit available
blocks
still falling-off-boundary, go to (3); Lena 409 752 39.2 608 704 44.3 0
3) Mark this block as vertical-not-available blocks. Peppers 397 850 39.8 606 673 43.9 0
Airplane 397 912 42.2 612 644 43.7 0
Different with [4], this paper select another method that Baboon 457 168 37.0 678 954 40.9 9
selects the last 2x rows to embed information about the Couple 412 824 40.2 611 308 43.6 25
vertical-not-available blocks. The information format is in Boat 421 080 38.9 617 789 42.8 0
Elaine 408 592 41.9 600 296 45.6 0
the order as: the Value of x (x is decided by total number of Man 423 560 39.1 624 027 43.5 7
blocks which are not available for embedding), the total
number of vertical blocks which are not available for
embedding, the first label of vertical block which is not TABLE II. COMPARISON OF PAPER [4] WITH THE PROPOSED METHOD
available for embedding, the second one’s label … until to Method in [4] Proposed method
the last one. The number of rows, the total number and each Cover image
Capacity/bit PSNR/dB Capacity/bit PSNR/dB
label is denoted by log 2 (nblocks ) bits. Using the modified 3- Lena 593 034 39.18 608 704 44.3
LSB algorithm shown in [5] to embed starting from the last Baboon 600 410 35.11 678 954 40.9
pixel of the image with reverse order. Airplane 590 800 38.05 612 644 43.7
Peppers 578 899 39.80 606 673 43.9
B. Extracting process
When extract the secret message, first of all, we extract As can be seen from Table 1, the number of vertical-not-
some information about the vertical-not-available blocks available blocks is very small, it means that it is reasonable
from last 2x rows, and then extract secret bits form the to use the last 2x rows to embedding information about
remaining 2 × 2 pixel blocks. The extraction process is as vertical-not-available blocks and usually x equals to 1. Table
follows: 1 and Table 2 show that, compare to [1], the embedding
(1). In the order of reverse to extract the lowest 3 bits capacity of our method is large than 50% and the value of
from last 2x rows. Take log 2 (nblocks ) as a basic unit, each PSNR is also large 3 dB~4 dB; compare to [4], the
unit’s mean is as the same as embedding process. For embedding capacity is similar the same, but our PSNR is
example, suppose that there is a image with 512×512, then average large than 5 dB.
the basic unit contain 16 bits, if the extracted bits is
000000000000000100000000000000010000010100001000. Pixel difference histogram analysis is shown in Figure 2.

283
 Number of pixels Original image Number of pixels Method in [1] Figure 4 is a security experiment over the proposed
15000 15000 method with the steganalysis method RS [7]. We can see
10000 10000 from the figure that Rm ≅ R− m , Sm ≅ S− m ˈit means the RS
method failed to detect the stego image, so that our method is
5000 5000
security against RS.
0 0
-100 -50 0 50 100 -100 -50 0 50 100 Value

Number of pixels Method in [4] Number of pixels Proposed method V. CONCLUSION

15000 20000

15000
This paper presents an image steganography method
10000 which utilize horizontal pixels and vertical pixel difference,
10000
in the horizontal direction we use high-quality model
5000 5000 function method for two pairs of pixels to embed message
0 0 and in the vertical direction we selective one pair of pixels to
-100 -50 0 50 100 -100 -50 0 50 100 Value
embed message with the common PVD methods.
Figure 2. Pixel difference histogram analysis Experiments show that this scheme has not only improved
the embedding capacity, stego image quality, but also
achieved high security such as resistance to common attacks
and RS histogram attack.
ACKNOWLEDGMENT
This paper was supported by the National High
Technology Research and Development Program of China
(a) Cover Image (b)Stego Image (863 Program) No. 2007AA01Z429; Natural Science
Figure 3. Visual comparison of cover image and stego image Foundation funding issues of Shaanxi Province No.
2010JM8034˗Foundation funding research project of the
Engineering College of Chinese Armed Police Force No.
Also we can see from Figure 2, the steganalysis method wjy201027.
shown in [3] can detect [1] and [4] easily because their
histograms have obvious steps, on the other hand, our REFERENCES
method’s histogram is the same as the histogram of original [1] Wu Da-chun, Tsai Wen-Hsiang. “A Steganographic Method for
cover image. Figure 3 is the comparison of image lena that Images by Pixel Value Differencing,” Pattern Recognition Letters,
2003, 24(9): 1613-1626.
before and after embedding message. Human eyes can’t
[2] Wang Chung-Ming, Wu Nan-i, Tsai Chwei-shyong, et al. “A High
distinguish it, it means that the stego image has good visual Quality Steganographic Method with Pixel Value Differencing and
quality. Modulus Function,” Journal of Systems and Software, 2008, 81(1):
150-158.
[3] Zhang Xinpeng, Wang Shuozhong. “Vulnerability of Pixel Value
Differencing Steganography to Histogram Analysis and Modification
for Enhanced Security,” Pattern Recognition Letters, 2004, 25(3):
331-339.
[4] BU Long, HU Bo. “An improved steganographic method by pixel-
value differencing,” Journal of circuits and systems, 2006, 11(3): 9-
13.
[5] Chan Chi-Kwong, Chen Lee-ming. “Hiding Data in Images by
Simple LSB Substitution,” Pattern Recognition, 2004, 37(3): 469-
474.
[6] Zheng jiangyun, Jiang julang. “Image quality assessment algorithm
based on sense capacity,” Computer engineering, , 2010, 36(8): 222-
223.
[7] Fridrich J, Goljan M, Du R. “Reliable Detection of LSB
Steganography in grayscale and color image,” Proc. ACM Workshop
Figure 4. RS steganalysis over the proposed method on Multimedia and security, 2001. 27-30.

284