You are on page 1of 4

A Steganographic Method with High Embedding Capacity by Improving Exploiting Modification Direction

Chin-Feng Lee 1, Yi-Ren Wang 2, and Chin-Chen Chang 3 1 Department of Information Management, Chaoyang University of Technology, 168, Jifong E. Rd., Wufong Township, Taichung County 41349, Taiwan, R.O.C. E-mail: lcf@cyut.edu.tw
2,3

Department of Computer Science and Information Engineering, National Chung Cheng University, 160 San-Hsing, Ming-Hsiung, Chiayi 621, Taiwan, R.O.C. E-mail: {ccc, wyj95m}@cs.ccu.edu.tw

Department of Information Engineering and Computer Science, Feng Chia University, 100 Wenhwa Rd., Seatwen, Taichung 40724, Taiwan, R.O.C. E-mail: ccc@cs.ccu.edu.tw Abstract
The stego image is the one that contains hidden secret messages. For a grayscale image, each pixel can be represented by an 8-bit integer, ranging from 0 to 255. Lots of existing LSB-based schemes embed secret messages by replacing the least significant bits of the cover pixels [2, 6, 7]. Chang et al. [1] proposed a method with high PSNR value by modifying the least significant bit of each pixel. Mielikainen [4] proposed an LSB revised method by changing a smaller range of pixel values than the traditional LSB does. Though modifying the LSBs of cover pixels can lower the distortion of stego image, the distortion-caused difference can still be readily detected by humans eyes or programs. Another key point for hiding secret message is the embedding capacity. The embedding capacity is defined as a ratio between the number of secret data bits and cover pixels. If the ratio is more than one, it represents the steganographic embedding scheme with a high embedding capacity [3]. For the scheme proposed by Tseng et al. [5], log 2 ( MN + 1) bits of secret data can be embedded into a stego image with size of M*N. Recently, Zhang et al. [8] proposed a steganography to exploit the modification directions called EMD embedding method, used to transform the binary secret data into the secret digits in the (2n+1)-ary notational system such that n pixels can be used to carry one secret digit. Zhang et al. showed that

We propose an effective data hiding method based on the EMD embedding scheme proposed by Zhang et al. in order to enhance the embedding capacity. The main idea of our proposed scheme is to convert secret binary messages into a sequence of digits in an 8-ary notational system. Each secret digit is embedded into two cover pixels by modifying the least significant bits (LSBs). As the experimental results show, the embedding bit-rate of our scheme is 1.5 times better than the EMD embedding scheme. In a word, we propose a high-quality and high-capacity data hiding method that can ensure the stego-image quality and security. Index TermsEMD embedding scheme, embedding capacity, embedding bit-rate, data hiding

1. Introduction
Steganography covers a carrier message for the sake of secure communication. During the past years, steganography has been used to embed secret messages into some objects, such as pictures, digital images and so on. Nowadays, steganography plays an important part for communication through Internet and becomes an extremely effective way to transmit secret messages.

their proposed scheme could achieve embedding efficiency and secrecy with low distortion. This paper revises the EMD embedding method to improve the embedding capacity. Compared with EMD embedding method, the embedding capacity of our method is 1.5 times, without losing the stego-image quality and security. The main idea of the proposed scheme is to convert a secret binary message into a sequence of digits in the 8-ary notational system. Each secret digit in the 8-ary notation system is carried by 2 cover pixels. Experimental results show that our proposed scheme still keeps the high quality of the stego image, while embedding capacity is enlarged.

2. The EMD embedding scheme


Zhang et al. [8] proposed an embedding method to exploit the modification directions. According to this EMD embedding method, one secret digit in the (2n+1)-ary notational system is carried by n cover pixels. The EMD embedding method using a secret key to pseudo-randomly permute all pixels in a cover image and partitions these permuted cover pixels into a series of groups. n grayscale pixels in a group (also called pixel-group) can be denoted as g1, g2, , gn. Before embedding, a data-hider should firstly convert secret messages into a sequence of secret digits in (2n+1)-ary notational system. Each secret digit falls within [0, 2n]. For each pixel-group, the extraction function f is defined as weighted sum function modulo (2n+1). f ( g1 , g 2 ,..., g n ) = ( g1 1 + g 2 2 + ... + g n n) mod(2n + 1). (1) If a secret digit to be embedded into a given cover pixel-group ( g1 , g 2 ,..., g n ) is not equal to the value calculated from the extraction function, only one of the cover pixels has to increase or decrease by 1; otherwise, none of the cover pixels can be modified. In an extracting procedure, if the stego pixel-group ' ' is ( g1' , g 2 ,..., g n ) , the extracting secret digit is the value of the extraction function. ' ' ' ' f ( g1' , g 2 ,..., g n ) = ( g1' 1 + g 2 2 + ... + g n n) mod n. (2) The EMD embedding scheme is a great contribution towards steganography. First, the PSNR value is greater than fifty. Although, the EMD embedding scheme has high quality of stego-image, only log 2 (2n + 1) binary bits are carried by n cover pixels. There is still for improving the embedding capacity.

decreasing function. The embedding bit-rate in 5-ary notational system, with one binary secret bit carried by one cover pixel, is the highest. However, we find out that more secret messages, in fact, can be hidden further. Thus we propose the high-quality and high-capacity data hiding method for improving the EMD embedding scheme. In our method, we transform a binary secret message into a sequence of secret digits in 8-ary notational system. A digit in 8-ary notational system can be completely transformed into 3-bit value in a binary system. We usually use dk to stand for a digit d in k-ary notational system. Two cover pixels in a pixel-group can be denoted as X and Y, where X and Y individually stand for the grayscale values in a cover image. In our proposed scheme, the extraction function fe is defined as the sum of the first pixel in pixel-group multiplied by one and another pixel multiplied by three modulo eight. (3) f e ( X , Y ) = ( X 1 + Y 3) mod 8. The extraction function f e proposed in our scheme is different from the extraction function f proposed in the EMD embedding scheme [8]. The hiding procedure for embedding secret data is listed as follows: Step1: Permute all pixels into a series of pixel-groups. Step2: For each pair (X, Y), perform the extraction function f e . Step3: For each embedding secret digit s, (3.1) If s is equal to fe(X,Y) , the pair of pixels (X, Y) does not need to be modified. (3.2) If s = fe (X +1Y) , X= X + 1. , (3.3) If s = fe (X 1Y) , X= X - 1. , (3.4) If s = fe (X,Y +1 , Y= Y + 1. ) (3.5) If s = fe (X,Y 1 , Y= Y - 1. ) (3.6) If s = fe (X +1Y +1 , X= X + 1 and Y= Y + 1. , ) (3.7) If s = fe (X +1Y 1 , X= X + 1 and Y= Y - 1. , ) (3.8) If s = fe (X 1Y +1 , X= X - 1 and Y= Y + 1. , ) The secret message extraction is performed as follows. Step1: Permute all stego pixels into a series of pixel pairs. Step2: For each pair of stego pixels ( X ' , Y ' ) , computes the secret digit F in the 8-ary notational system by the following extraction function f e .

3. Our proposed scheme


According to the EMD scheme proposed in [8], the embedding bit-rate function f (n) = log 2 (2n + 1) is a
n

F ' = f e ( X ' , Y ' ) = ( X ' 1 + Y ' 3) mod 8.

(4)

Step3: Convert the sequence of secret digits into its corresponding binary secret message. For Eq. (3), to prevent the overflow and underflow problems, i.e., to let the first and second pixels be in the range of [0, 255], we have to change the value of cover pixel before embedding secret data.

Though the experimental results of Tables 6 and 7 show that the PSNR values of our method are slightly lower than those of the EMD embedding scheme, the differences between both methods are extremely trifling and minute in human visual perception. Even so, our proposed scheme still outperforms than the EMD embedding method. Table 1. The maximum embedding capacities by the EMD embedding method under 5-, 7-, and 9-ary notation systems and our proposed scheme

4. The experimental results


We test the proposed algorithm on various grayscale images including Lena, Pepper, Baboon, Boat, Babala, and Goldhill. These test images are of size 256 256 or 512 512. The embedded payload sizes, bit-rates and peak signal to noise ratios (PSNR) of embedded images for test images are listed in Tables 1-5, respectively. In Tables 1, 4, and 6 the size of each test image is 256*256. The size of each test image in Tables 2, 5, and 7 is 512*512. For each test image, we conduct some experiments to compare the embedding capacity between our proposed scheme and Zhang et al.s EMD embedding method [8]. The experimental results of the highest embedding capacity and embedding bit-rate are shown in Tables 1, 2 and 3. In Tables 1 and 2, we define that the difference of the embedding capacity between our method and Zhang et al.s as gain in bits. The gain is used to show the increased number of bits of the embedding capacity of our method comparing with that of Zhang et al.s. From Tables 1 and 2, we can easily find that the gains increase as the test image sizes increase. Since, the embedding bit-rate function log 2 (2n + 1) is decreasing, the maximum f ( n) =
n

Table 2. The maximum embedding capacities by the EMD


embedding method under 5-, 7-, and 9-ary notation systems and our proposed scheme

Table 3. The maximum embedding bit-rates by the EMD


embedding method under 5-, 7-, and 9-ary notational systems and our proposed scheme

embedding capacity appears when the EMD embedding method is adopted under the situation that the secret digit in the 5-ary notational system is carried by two cover pixels. This means that, larger n will lead to less embedding capacity for a (2n+1)-ary notational system using the EMD embedding method. The embedding bit-rates in our scheme remain unchanged and is higher than that of EMD method as shown in Figure 1. From Tables 1, 2 and 3, the embedding capacity and bit-rate of the EMD embedding method are 65536, 262144 and one, respectively. These figures are apparently the highest under the 5-ary notation system no matter which test image is adopted. Our proposed scheme uses the 8-ary notational system to convert the secret messages and develops a new extraction function such that the embedding bit-rate of our proposed scheme is more than 1.5 times as that of Zhang et al.s method [8].

Figure 1. The difference of the maximum embedding


bit-rates between the EMD embedding method and our proposed scheme

Table 4. The PSNR values from various test images with sizes of 256*256 by our proposed scheme

5. Conclusions
In this paper, we propose a high-quality and high-capacity data hiding method by improving the EMD embedding scheme to enhance hiding capacity. We convert the secret binary messages into a sequence of digits in the 8-ary notational system before hiding the secret messages. In the proposed scheme, one secret digit is carried by two cover pixels. In addition, we develop a new extraction function which is the sum of the first pixel in two-pixel pair multiplied by one and another pixel multiplied by three modulo eight. If the value of extraction function is equal to the secret digit, the pair of cover pixels is not modified. Otherwise, if the value of extraction function is not equal to the secret digit, we should change the value of cover pixels, increasing or decreasing by 1, in order to let the result of the extraction function for the pair of pixels equal to the secret digit. This makes our scheme suitable for increasing the embedding capacity and keeping good quality.

Table 5. The PSNR values from various test images with sizes of 512*512 by our proposed scheme

Table 6. The differences between our proposed scheme and the EMD embedding method under 5-, 7-, 8-, and 9-ary notational systems (test image size: 256*256)

References
1 C. C. Chang and W. C. Wu, A Novel Data Hiding Scheme for Keeping High Stego-Image Quality, Proceedings of the 12th International Conference on MultiMedia Modelling, Bijing, China, January 2006, pp. 225-232. 2 A. Ker, Steganalysis of LSB Matching in Grayscale Images, IEEE Signal Processing Letters, vol. 12, no. 6, June 2005, pp. 441-444. 3 Y. K. Lee and L. H. Chen, High Capacity Steganographic Model, IEE Proceedings-Vision, Image and Signal Processing, vol. 147, no. 3, 2000, pp. 288-294. 4 J. Mielikainen, LSB Matching Revisited, IEEE Signal Processing Letters, vol. 13 no. 5, May 2006, pp. 285-287. 5 Y. C. Tseng, Y. Y. Chen, and H. K. Pan, A Secure Data Hiding Scheme for Binary Images, IEEE Transactions on Communications, vol. 50, no. 8, August 2002, pp. 1227-1231. 6 R. Z. Wang, C. F. Lin, and J.C. Lin, Image Hiding by Optimal LSB Substitution and Genetic Algorithm, Pattern Recognition, vol. 34, no. 3, 2001, pp. 671-683. 7 H. C. Wu, N. I. Wu, C.S. Tsai, and M.S. Hwang, Image Steganographic Scheme Based on Pixel-Value Differencing and LSB Replacement Methods, IEE Proceedings-Vision, Image and Signal Processing, vol. 152, no. 5, October 2005, pp.611-615. 8 X. Zhang and S. Wang, Efficient Steganographic Embedding by Exploiting Modification Direction, IEEE Communications Letters, vol. 10, no. 11, November 2006, pp. 1-3.

Table 7. The differences between our proposed scheme and


the EMD embedding method under 5-, 7-, 8-, and 9-ary notational systems (test image size: 256*256)

You might also like