Abstract: Information technologies and communications have pervaded our homes and business places. No matter how well-organized and extensive the communication technology is there are always loop holes in the network and people who seek after the clandestine information to extract from these loop holes. The pandemic problem of security is still a raging problem as every solution compromises on one trait to heighten the other(s) according to the necessity of the hour. In this paper we suggest an algorithm where we have tried to retain all the three important banalities of secure communication: robustness, capacity and imperceptibility by using Haar Integer wavelet transform a Discrete Wavelet Transform (DWT) domain method in tandem with a cryptic scheme i.e. the direct binary or inverted binary embedding of data. Experimental results that compute the MSE and PSNR show that this algorithm caters to the need of the hour by delivering the high capacity with good imperceptibility.
INTRODUCTION
In this age of booming communication and technology, everything is just one phone call or e mail or a flight away. The free flow of information has behooved the need for protecting this huge wealth of knowledge that has from time to time helped tremendously in growth of science, technologies, civilizations and mankind as a whole. With the invention of various algorithms for intelligent storing, processing and transmitting of information ranging from minutiae procedures to long exhaustive routines, many techniques have been developed for computer security, information security and information assurance. The most prominent amongst them is steganography (Stefan and Fabin, 2000, Petitcolas et al., 1999; Rabah, 2004; Cheddad et al., 2010). It involves communicating secret data in an appropriate multimedia carrier (Bender et al., 1996; Rabah, 2004), e.g., image (Hmood et al., 2010a; Amirtharajan and Balaguru, 2009; Amirtharajan et al., 2011), audio (Zhu et al., 2011), video (Al-Frajat et al., 2010) and text files (Al-Azawi and Fadhil, 2010; Yang et al., 2011; Shirali-Shahreza and Shirali-Shahreza, 2008).
Cryptography another popular technique for information security involves scrambling of data in an unintended format which would seem gibberish to any unintended user (Schneier, 2007). No extraction can be done unless the third party knows the secret code. However, it would invite tampering just because of the existence of a secret message. That’s where steganography has gained its impetus, with its basic constituents being a cover image, secret data and a key (Stefan and Fabin, 2000; Petitcolas et al., 1999; Rabah, 2004). The basic differences between cryptography and steganography is given by Zaidan et al. (2010) and Cheddad et al. (2010). When combined it gives out a stego image which has high un-detectability, is robust and also has high capacity (Hmood et al., 2010b) fulfilling requirements of the “Fridrich’s magic Triangle” in information hiding (Stefan and Fabin, 2000; Padmaa et al., 2011). Any files such as audio, video or image can be used as cover (Bender et al., 1996; Rabah, 2004), cover being the carrier without secret data in it, after embedding secret data in the cover; it is camouflaged with high level of imperceptibility and dexterity to yield the stego image. Here capacity or payload would mean the total amount of secret information that can be hidden in the stego image, without any physical notice-ability in the characteristics of the cover image while robustness defines the limit of standing modifications before an adversary can destroy the stego image. Within steganography, there are various techniques or methods to embed the data in the cover image i.e., Substitution based (Amirtharajan and Balaguru, 2009; Chan and Cheng, 2004), Transform domain based (Thanikaiselvan et al., 2011a), Spread Spectrum based (Amirtharajan and Balaguru, 2011; Thenmozhi et al., 2011; Kumar et al., 2011), Statistics based (Qin et al., 2009; Zanganeh and Ibrahim, 2011), Distortion and cover generation based (Xiang et al., 2011; Stefan and Fabin, 2000).
The most prominent being Spatial Domain based Steganographic Techniques and Transform Domain based Steganographic Techniques (Thanikaiselvan et al., 2011a; Kumar et al., 2011), Spatial domain based steganography include the Least Significant Bit (LSB) technique (Amirtharajan and Balaguru, 2009; Chan and Cheng, 2004), Pixel value differencing (Amirtharajan et al., 2010; Thanikaiselvan et al., 2011b; Padmaa et al., 2011) and more while the latter includes DCT (Provos and Honeyman, 2003), DWT and especially IWT (Thanikaiselvan et al., 2011a; EI Safy et al., 2009). A detailed survey on steganography till 1999 is available in Petitcolas et al. (1999) whereas a detailed survey on Information hiding in images, differences among cryptography, Stegano graphy (Zaidan et al., 2010) and water marking is available Cheddad et al. (2010). The counter attack called steganalysis is described by Qin et al. (2009) and Xia et al. (2009) and various steganalysis review is detailed by Qin et al. (2010).
A steganography technique is considered to be dependable when it can retain the embedded data in spite of severe modifications done to it and not displaying the payload contained in it. The most employed technique for this involves transforming the cover image into another domain and embedding data in the transformed pixels (Thanikaiselvan et al., 2011a; EI Safy et al., 2009). Before transformation the cover image exists in spatial domain, which is transformed into the frequency or time domain for the coefficients and after embedding data in the transformed pixels, it is brought back to the spatial domain. Thus, the underlying idea is, even if image is subjected to modifications and is in worst cases transformed, the data will still be hidden in the transformed pixels (Thanikaiselvan et al., 2011a).
Transformation techniques include DCT, DWT and IWT, though DCT isn’t highly preferred as it has low hiding capacity (Provos and Honeyman, 2003). Contemporary researchers use DWT, since it can be employed in compression of formats JPEG2000 and MPEG4. Techniques that use DWT i.e., wavelet transform based stego technique provides high capacity with the secret message embedded into the high frequency and low frequency coefficients of the wavelet transform, but provides less PSNR at a high hiding rate.
Cryptography in coalesce with steganography either through random walk (Luo et al., 2008) or variable bit optimal embedding (Zanganeh and Ibrahim, 2011) could be an effective solution to improve the complexity. In this study we propose a new modified version of the methodology by Thanikaiselvan et al. (2011a) which can embed a larger amount of data in Integer Wavelet Transform (IWT) domain with high PSNR while combining with the inverted pattern approach (Yang, 2008; Amirtharajan and Rayappan, 2012) to improve the complexity.
Related works: The use of Wavelet transform will mainly address the capacity and robustness of the information- hiding system features. The Haar Wavelet Transform is the simplest of all wavelet transform. In this the low frequency wavelet coefficients are generated by averaging the two pixel values and high frequency coefficients that are generated by taking half of the difference of the same two pixels (EI Safy et al., 2009). The four bands obtained are LL, LH, HL and HH which is shown in Fig. 1. The LL band is called as approximation band which consists of low frequency wavelet coefficients and contains significant part of the spatial domain image. The other bands are called as detail bands which consist of high frequency coefficients and contain the edge details of the spatial domain image.
Integer wavelet transform: Integer wavelet transform can be obtained through lifting scheme. Lifting scheme is a technique to convert DWT coefficients to Integer coefficients without losing information.
Forward Lifting scheme in IWT:
Step 1: Column wise processing to get H and L:
H = (Co-Ce) |
(1) |
L = (Ce-[H/2]) |
(2) |
| Fig. 1: | Image and its transform domain bands |
Where, Co and Ce is the odd column and even column wise pixel values
Step 2: Row wise processing to get LL, LH, HL and HH, Separate odd and even rows of H and L, Namely
LH = Lodd-Leven |
(3) |
LL = Leven-⌊LH/2⌋ |
(4) |
HL = HoddHeven |
(5) |
HH = Heven-⌊HL/2⌋ |
(6) |
| Where: | ||
| Hodd | = | Odd row of H |
| Lodd | = | Odd row of L |
| Heven | = | Even row of H |
| Leven | = | Even row of L |
Reverse lifting scheme in IWT: Inverse Integer wavelet transform is formed by Reverse lifting scheme. Procedure is similar to the forward lifting scheme (Thanikaiselvan et al., 2011a; EI Safy et al., 2009).
LSB Embedding: Simple LSB embedding is detailed by Amirtharajan and Balaguru (2009), Chan and Cheng (2004) and Janakiraman et al. (2012). Consider a 8-bit gray scale image matrix consisting mxn pixels and a secret message consisting of k bits. The first bit of message is embedded into the LSB of the first pixel and the second bit of message is embedded into the second pixel and so on. The resultant stego-image which holds the secret message is also a 8-bit gray scale image and difference between the cover image and the stego-image is not visually perceptible. This can be further extended and any number of LSB’s can be modified in a pixel. The quality of the image, however degrades with the increase in number of LSB’s. Usually up to 4 LSB’s can be modified without significant degradation in the message. Mathematically, the pixel value ‘P’ of the chosen pixel for storing the k-bit message Mk is modified to form the stego-pixel ‘Ps’ as follows:
Ps = P-mod (P,2k)+Mk |
(7) |
The embedded message bits can be recovered by:
Mk = mod (Ps,2k) |
(8) |
One method to improve the quality of the LSB substitution is Optimal Pixel adjustment Process (OPAP) (Chan and Cheng, 2004).
Determination of embedding:
| • | Direct Binary embedding of data is done and the MSE (Mean Square Error) calculated |
| • | Let it assumed to be X |
| • | Inverted Binary embedding of data is done and the MSE (Mean Square Error) calculated |
| • | Let it assumed to be Y |
| • | Depending upon values of X and Y, If X > Y |
| • | Then the value is determined to be ‘1’ and if Y > X |
| • | Then the value is determined to be ‘0’ |
PROPOSED METHODOLOGY
The proposed block diagram of high capacity steganography system is given in Fig. 2 and 3. Preprocessing includes R, G and B plane separation and Histogram modification. Then Integer wavelet transform is applied to the cover image to get wavelet coefficients. Wavelet coefficients are randomly selected by using key-2 for embedding the secret data. Key -2 is 8x8 binary matrix in which ‘1’ represents data embedded in the corresponding wavelet coefficients and ‘0’ represents no data present in the wavelet coefficients. Then the direct binary or inverted binary embedding of data is done in the respective co-efficient. Key-1(K1) is a decimal number varying from 1 to 4 and it will decide the number of bits to be embedded in the cover object. High capacity is achieved by varying the key-1 (K1) value.
| Embedding Algorithm | |
| Step 1: | The cover image of size 256x256x3 pixels is selected |
| Step 2: | The respective planes are separated into R, G and B constituents |
| Step 3: | The required data file to be embedded is taken with each character taking 8 bits |
| Step 4: | Histogram modification is done in all planes, Because, the secret data is to be embedded in all the planes, while embedding integer wavelet coefficients produce stego-image pixel values greater than 255 or lesser than 0. Then all the pixel values will be ranged from 15 to 240 |
| Step 5: | Each plane is divided into 8x8 blocks |
| Step 6: | Apply Haar Integer wavelet transform to 8x8 blocks of all the planes, this process results in LL1, LH1, HL1 and HH1 sub bands |
| Step 7: | Using Key-1(K1) calculate the Bit length (BL) for corresponding wavelet co-efficients (WC), here we used modified version of Bit length calculation used by Thanikaiselvan et al. (2011a). Using the following equation, we get the high capacity steganography |
| (9) |
| Step 8: | Using key-2 select the position and coefficients for embedding the ‘BL’ length data using LSB substitution. Here data is embedded only in LH1, HL1and HH1 subbands. Data is not embedded in LL1 because they are highly sensitive and also to maintain good visual quality after embedding data. An example of key-2 is shown below (This is Key B) |
| (10) |
| Step 10: | After doing required operations with K-1 and K-2, the direct binary embedding of required data is done in the determine position and co-efficient and MSE calculated |
| Step 11: | Next inverted binary embedding of the same data is done in the same position and corresponding MSE calculated |
| Step 12: | Depending upon MSE values obtained Key-3 (K-3) is obtained with value ‘1‘ when inverted binary is embedded and ‘0‘ when direct binary is embedded. Thus K-3 is also an 8*8 matrix consisting of 1s and 0s representing whether direct or inverted binary embedding of data is done |
| Step 13: | Applying Optimal Pixel adjustment Procedure (OPAP) reduces the error caused by the LSB substitution method |
| Step 14: | Take inverse wavelet transform to each 8x8 block and combine R,G&B plane to produce stego image |
| Extraction Algorithm | |
| Step 1: | The corresponding stego image of 256*256 pixels is selected |
| Step 2: | The respective planes are separated into R, G and B constituents |
| Step 3: | Each plane is divided into 8x8 blocks |
| Step 4: | Apply Haar Integer wavelet transform to 8x8 blocks of all the planes, This process results LL1,LH1, HL1 and HH1 sub-bands |
| Step 5: | Using Key-1 calculate the Bit length(BL) for corresponding wavelet co-efficients(WC), using the ‘BL’ equation used in Embedding procedure |
| Step 6: | Using key-2 select the position and coefficients for extracting the ‘BL’ length data |
| Step 7: | Then using key-3, determine whether direct or inverted binary of the data has been embedded and extract it then depending upon it |
| Step 8: | Combine all the bits and divide it in to 8 bits to get the text message |
| Fig. 2: | Block diagram for Embedding |
| Fig. 3: | Block diagram for extraction |
ERROR METRICS
A performance measure in the stego image is measured by means of two parameters namely, Mean Square Error (MSE) and Peak Signal to Noise Ratio (PSNR). The MSE is calculated by using the equation:
| (11) |
where, M and N denote the total number of pixels in the horizontal and the vertical dimensions of the image Xi, j represents the pixels in the original image and Yi, j, represents the pixels of the stego-image. The Peak Signal to Noise Ratio (PSNR) is expressed as:
| (12) |
RESULTS AND DISCUSSION
In this present implementation, Lena and baboon 256x256x3 color digital images have been taken as cover images, as shown in Fig. 4a and 5a, Stego lena (Thanikaiselvan et al., 2011a) in Fig. 4b and 5b and Proposed Stego images in Fig. 4c and 5c and tested with key-1(key-J) and various key-2s.
| Fig. 4(a-c): | (a) Cover Images Lena; (b) Stego lena (Thanikaiselvan et al., 2011a); (c) Proposed Stego lena |
The effectiveness of the stego process proposed has been studied by calculating MSE and PSNR for the Lena and Baboon digital image in RGB planes and tabulated.
In this analysis, Key-2 is used for random selection of coefficients for embedding data (in this analysis Key-1 has been set as K1 = 1) and the results are tabulated in Table 1 for various Key-2 using the proposed method.
| Fig. 5:(a-c): | (a) Cover Images Baboon; (b) Stego Baboon (Thanikaiselvan et al., 2011a); (c) Proposed Stego Baboon |
| Table 1: | Comparative Analysis for the proposed method key - 1(K1) = 1 |
It’s evident that from Table 1, Key-J provides high capacity and Key-A provides low capacity, moreover proposed methodology gives High PSNR over previous method (Thanikaiselvan et al., 2011a). By keeping Key J constant various key 1 are tabulated in Table 2.
| Table 2: | Key J constant by varying Key 1 for Lena and Baboon |
| Method* -Thanikaiselvan et al. (2011a) | |
Complexity level estimation: For 8x8 pixels block case IWT, total number of blocks (N) = 1024 Number of PRNG output for randomizing the 1024 blocks:
= NpR = N!/(N-r)! = 1024p1024 = 1024! |
As per DES, the complexity of each block = 2^64.
Either binary or inverted binary would be embedded hence complexity increases
by 2 and 3 out 4 sub bands are used and based on key 2 either 0 or 1 then 0.5.
Therefore the total complexity:
= (2^64)*(1024)! * (2) *3/4 *(.5) |
These embedding are carried out in transform domain and the security level estimation reveals the firmness of the proposed stego against hackers.
CONCLUSION
It has been observed that steganography provides excellent avenue for high payload combined with imperceptibility. Literature suggests that in several techniques robustness may have been compromised, however the proposed method gives high payload (capacity) in the cover image with very little error. The algorithm is robust, as there is an option of embedding data in the transformed domain and also the option of reducing the error by determining whether direct or inverted binary maybe embedded. Key-1 and Key-2 not only provide high security but also increased capacity with the wavelet transform. The drawback of the proposed method is the computational overhead. This can be reduced by high speed computers. Thus, it can be summarized as:
| • | PSNR is increased in this system with intelligent use of Key-1, Key-2 and Key-3 |
| • | Because data is embedded in imperceptible areas that too in the transformed domain, the stego image can hardly raise suspicion |
| • | Even if the stego image is transformed it would be difficult to determine what data has been embedded without Key-3 as either direct or inverted binary would be embedded. Thus capacity, imperceptibility and robustness, all the three requirements are catered to in this innovative algorithm |