PVD Based Steganography on Scrambled RGB Cover Images with Pixel Indicator

With the rapid advancements in digital information transmission, technologies need for ensuring high level of data security, has become indispensible. In order to ensure the same, many cryptography and steganography techniques are in use. In the proposed methodology, both the techniques have been used simultaneously for incorporating a near fool proof security. Cryptography renders an image unreadable whereas, steganography aims at hiding secret information in the image. In the scrambled image, adaptive Least Significant Bit (LSB) substitution has been performed using Pixel Value Differencing (PVD) based on Pixel Indicator (PI) method for colour images. Here, secret data is embedded only in R and G planes while, B plane acts as an indicator for embedding. This methodology increases the embedding capacity by 50% compared to the existing methods in addition with a reasonable Peak Signal to Noise Ratio (PSNR) of 42 dB.

INTRODUCTION

With the advancements in information technology, issue of data security comes hand in hand. Large amount of data transmitted via internet needs to be made secure against malicious attacks. Secret information particularly sensitive information needs to be handled with extreme care as its misuse can create great havoc to the society. So, to make the secret information attack proof, role of various cryptographic and steganographic techniques come into picture (Cheddad et al., 2010; Amirtharajan and Rayappan, 2012a-d, 2013; Amirtharajan et al., 2013a-j; Janakiraman et al., 2012a, b, 2014a, b; Luo et al., 2011; Mohammad et al., 2011; Salem et al., 2011; Ramalingam et al., 2014a, b; Thien and Lin, 2003; Zhao and Luo, 2012).

Cryptography has been in practice from ancient times with its main aim being making the message unreadable by all except the authorised receiver who has the key to retain the actual message. It converts the secret information in un-intelligible form thereby, making it unreadable by a hacker. On the other hand, steganography aims at hiding data in multimedia such as image (Chan and Cheng, 2004; Amirtharajan et al., 2010, 2011, 2012, 2013b), audio or video files, so that secret data is invisible to the naked eye during transmission called information security (Thenmozhi et al., 2012; Praveenkumar et al., 2012a, b, 2013a, b, 2014a-j). The ideology behind this is, if a certain feature of hidden information becomes visible, point of attack becomes evident. Various algorithms and methodologies exist for the same in hardware (Rajagopalan et al., 2012a, b, 2014a-d; Janakiraman et al., 2012a, b, 2014a, b).

Three main pillars of steganography are ensuring robustness (Wong et al., 2007; Qi and Wong, 2005), high embedding capacity and undetectability (Wu and Tsai, 2003; Zhang and Wang, 2004; Thanikaiselvan et al., 2012a-c, 2013a, b). This study proposes use of cryptography and steganography together for hiding text files in colour images which are used as cover. The RGB colour planes of the image are separated and scrambled, except the blue plane, using Arnold Cat map. Image scrambling technology is easy to realized but it is not in accordance with Kirchhoff’s rules and its security is not very high. But when data is embedded in this scrambled image using steganography algorithms, a high level of security is ensured.

Blue plane of the image is used as indicator for embedding in R and G planes where the clandestine context gets pushed in the adaptive LSB values so as difference in the pixel value is not visible to the naked eye. If the pixel value of first pixel in 2x2 matrix of blue plane is even, then data is embedded in red plane else, it is embedded in green plane. Data embedding takes place by Pixel Value Differencing (PVD) method where four pixels of the 2x2 matrix of the cover used at once for embedding the data (Wu and Tsai, 2003; Zhang and Wang, 2004). Pixel indicator method (Janakiraman et al., 2012a, b, 2013) is a method which can be adapted with PVD and other methods to improve the security further. In this study, PVD and pixel indicator method are used on a scrambled cover image, data is embedded adaptively in R and G plane i.e., k bits are embedded based on the level inwhich the average difference falls based on the pixel value in B plane. This technique improvises on pixel embedding capacity as well as ensuring a high level of security.

METHODOLOGY

The methods publicized here put forward a secure data hiding technique by using both cryptography and steganography techniques. The block diagram for the method is shown in Fig. 1. This technique is aimed to achieve high security by first scrambling the colour planes of cover file which is a 24 bit colour image using Arnold algorithm and then embedding the secret data in the encrypted R and G planes using a steganographic algorithm based on pixel indicator on the blue plane. For transmission, the R and G planes are descrambled and merged to form colour stego image and then transmitted thereby hiding the very essence of any data hidden. As the steganography is performed on the encrypted cover file, any attempt to search for hidden data in the transmitted file will result in garbage values. This ensures a highly secure way to transfer sensitive data as compared to use of cryptography and steganography alone.

At receiver’s end, the transmitted image undergoes the encrypting process again using the key. The secret data is then extracted from the image.

Image encryption algorithm using Arnold Transform method: The colour planes are separated and Arnold Transform is used as the encryption technique for scrambling pixels in these R and G planes. Arnold Transform is applied to the planes in the spatial domain itself. For encryption of the cover image, Eq. 1 is used:

(1)

where:

is the transfromation matrix and MxM pixels is the size of the cover image.

Fig. 1:	Proposed block diagram

Here, the dimension of the cover image is 512x512 pixels. Also, a, b are the pixel location of the original image, a', b' as a pixel location of the scrambled image. The location of one pixel is changed multiple times. The number of times it is changed is the key 1 for the encryption process and is stored as the number of iterations. The process is same for both the planes.

The Arnold Transform method of encryption is periodic in nature. It is robust, quick and has high confidentiality. After Arnold transformation of the image, the pixel’s location is transformed but this transformation will not change the values of the pixel values hence, the image histogram is the same.

Image steganography: Now, the R and G planes have been encrypted, the secret data is embedded in an order in the R and G plane depending on whether the left topmost corner pixel in the 2x2 matrix of blue plane is even or odd. The secret data can be a text file which has been converted into binary. In this study, only binary data has been embedded in R and G planes according to Table 1.

(2)

Secret (n-bit) is denoted by M as shown in Eq. 3:

Table 1:	Pixel indicator method

(3)

where, m-bit secret message has to be embedded into the k (k may be 1, 2, 3, 4) rightmost LSBs of the colour planes R and G. This m-bit binary message is converted to its decimal value by combining k bits together (Z_ip). The substitution is done by using the Eq. 4:

(4)

The new pixel value, a_ij’, has the secret data embedded in last k bits. As data gets infixed only in LSBs, the pixel value does not change significantly and is not visible to the human eye.

A threshold value Th is taken for the embedding process. The range of Th is 2^kl≤Th≤2^kh, where kl and kh are the number of bits embedded in the lower threshold and upper threshold, respectively.

A sub-block of 2x2 pixels is taken from the image. Out of the 4 pixels, the minimum (P_min) is found. Then using formula in Eq. 5, value of delta is estimated. According to the relation of Delta with Th, either higher level embedding (if Delta>Th) or lower level embedding (if Delta<Th) is performed, i.e., either kh or kl bits of data is embedded, respectively.

(5)

where, P₁, P₂, P₃, P₄ are the four pixel values of the 2x2 sub-block.

This method for steganography gives higher embedding capacity and better image quality. The secret data is secure as the key 2 (i.e., the threshold value Th, kl and kh) needs to be known for the extraction of information. Without the key, there is no way of extracting the data.

Decryption: Before the transmission of the stego image, the R and G planes are decrypted so as to get the original cover image. This is done so that the hackers do not suspect the presence of any information in the image. If the hackers try to get information then, it will result in garbage values as the decryption process has changed the sequence inwhich data was stored. Equation 6 is used for the decryption process; it is the inverse of the Arnold Transform Matrix:

(6)

where:

is the inverse transformation matrix and MxM is the cover’s dimension which is 512x512. Also, a, b are the pixel location of the scrambled image, a', b' as a pixel location of the descrambled image. The multiplication i.e., pixel location change is performed as many times as when it was scrambled. The decrypted image is now ready for transmission and the colour planes are merged after descrambling.

Stego extraction: The first step of the stego extraction process is to scramble the individual R and G planes again using Arnold Transform and the key 1 as described above. The key 2 contains the information about the Threshold Th and the kl and kh, i.e., bits inserted at different levels. A sub-block of 2x2 pixels is taken from the R, G and B planes. Out of the 4 pixels, find the minimum pixel value for the R and G planes and find the delta value using Eq. 5, separately for the R and G planes. If the remainder after dividing the left topmost pixel in sub-block in B plane be 0, then bits are primarily extracted from the R plane followed by the G plane. On the other hand, if the remainder is 1, bits are first extracted from the G plane and then from R plane. According to the relation of the individual delta of the R and G planes with THz, either kh bits of information (if Delta>Th) or kl bits (if Delta<Th) is extracted. Thus, the information embedded has been successfully extracted.

Thus, the binary information that was embedded has been successfully extracted. Now, convert the binary information to the string format.

RESULTS AND DISCUSSION

Experiments using Matlab software have been performed on 5 colour images with size 512x512. PSNR values for each red and Green plane along with number of bits embedded in each scrambled image has been calculated. A text file has been used as surreptitious content that gets rehabilitated from its ASCII value to binary to make it suitable for embedding in the scrambled cover image. The PSNR is used for estimating scrambled stego image’s quality. For a PxQ gray scale image, PSNR is calculated as shown in Eq. 7:

(7)

where, a_i,j pixels in ith row and jth column of cover whereas, b_i,j is the same as that of stego.

Stego images generated for this proficiency are shown from Fig. 2-7. Figure 2 shows five colour images of size 512x512 which are the original images. Figure 3 and 4 shows the result after scrambling the R and G plane using Arnold Algorithm. The scrambling can be performed n number of times. Figure 5 and 6 shows stego R and G images with data embedded inside using PVD and LSB substitution. Figure 7 shows descrambled stego colour images obtained after merging the RGB planes transmitted through a suitable channel. As can be seen, no significant difference can be noticed in original images and descrambled stego images.

Table 2 contains the capacity and the PSNR for five different images for varied threshold and levels of embedding.

Fig. 2(a-e):	Cover images of size 512x512, (a) Peppers, (b) Baboon, (c) Castle, (d) Barbara and (e) Boat

Fig. 3(a-e):	R plane scrambled images with size 512×512, (a) Peppers, (b) Baboon, (c) Castle, (d) Barbara and (e) Boat

Fig. 4(a-e):	G plane scrambled images of size 512×512, (a) Peppers, (b) Baboon, (c) Castle, (d) Barbara and (e) Boat

Fig. 5(a-e):	Stego.R plane scrambled images of size 512×512, (a) Peppers, (b) Baboon, (c) Castle, (d) Barbara and (e) Boat

Fig. 6(a-e):	Stego. G plane scrambled images with size 512×512, (a) Peppers, (b) Baboon, (c) Castle, (d) Barbara and (e) Boat

Fig. 7(a-e):	Descrambled Stego colour images with size 512x512, (a) Peppers, (b) Baboon, (c) Castle, (d) Barbara and (e) Boat

In Table 2, various threshold values as Th = 5, 6, 12, 18 with various kl-kh values have been taken and PSNR value for Red and Green planes along with bit embedding capacity has been calculated. For example, at Th = 5, if Delta<=Th, 2 bits are embedded and at Delta>Th, 3 bits are embedded in a 4 pixel block. It can be observed that with the increase in threshold value, PSNR reduces but the embedding capacity (in bits) increases at the same time.

The comparative analysis for the difference in the levels of embedding when the scrambled R and G planes are used for embedding versus when original R and G planes are used is shown in Table 3. As can be seen, the capacity increases with the scrambling of R and G planes as compared to using original R and G planes.

Table 4 shows the comparative analysis between proposed and other spatial domain steganography methods. Scrambled cover image is well suitable for PVD based steganography because it offers high capacity and high PSNR over other methods. For the comparative analysis, a 256x256 RGB baboon image is taken as a cover image. In RGB images one pixel value takes 24 bits for showing all color combinations. In the proposed method, total 393158 bits are embedded in a single image with PSNR of 42.5 dB (average). This shows that the proposed method has high capacity with high imperceptibility than the other methods. Moreover, Red and Green planes are only used for embedding process and the Blue plane is used as an indicator for embedding process.

Table 2:	Capacity and PSNR values for various T, kl and kh

Table 3:	Comparative analysis between with and without scrambling cover image

Table 4:	Comparative analysis for determination of cover image of baboon between other spatial domain methods

CONCLUSION

This study recommends a steganographic data security algorithm for ensuring high level of data security. First, the Cover Image has been scrambled using Arnold algorithm and data has been embedded in the scrambled image to incorporate high degree of randomness using PVD and k-bit LSB substitution according to pixel indicator which is blue plane in this case. It has been observed that as the threshold value goes up, the embedding capacity increases significantly but there is a decrease in PSNR value that is the picture quality degrades. It is vivid that the embedding competence increases through the use of scrambling R and G planes as compared to using original R and G planes, in addition to increasing security. This scheme majors in the grounds of entrenching capability and making secret data fool proof. However, there is a tradeoff between PSNR and embedding capacity at high threshold values. Improving the same will be our future endeavor.