subong2018

2018 15th International Joint Conference on Computer Science and Software Engineering (JCSSE) LSB Rotation and Inversion Scoring Approach to Image Steganography Ryan A. Subong Arnel C. Fajardo Yoon Joong Kim Computer Engineering Department School of Engineering and IT Department of Computer Engineering Western Institute of Technology Manuel L. Quezon University Hanbat National University Iloilo City, Philippines Daejeon, Republic of Korea [email protected] Quezon City, Philippines [email protected] [email protected] Abstract— The less obvious that an image has been modified, Steganography De-steganography the less it will be suspected of containing a secret message or Process Process image. This paper proposes an image steganographic approach wherein the bit information of the secret message replaces the Cover image Stego image LSBs (least significant bit) of the RGB (red green blue) bytes of the cover image just like many of the LSB image steganography Secret image Extracted methods, except that the bits of the secret message undergo a Secret image series of evaluated and scored bit rotation and inversion operations prior replacement. Using MSE and PSNR as a Fig. 1. Steganography and de-steganography process measure of image quality, the stego image generated by this proposed approach produced lesser distortion than the existing II. EXISTING ALGORITHMS AND TESTS four bits per byte replacement approach of LSB Replacement and Adaptive LSB Embedding algorithms. The proposed The most commonly used image steganography method is approach however does not offer significant improvement of the LSB Replacement algorithm [5] in which the embedding of robustness in terms of security. the secret message is done only to the least significant bits of the cover image so as to minimize its distortion effect to the Keywords— Bit Rotation and Inversion; Cryptography; Image stego image. In using a 24-bit colored image as a cover image, Steganography; LSB Embedding; the process stores data bits of the secret message in the least significant bits of the RGB (red, green and blue) components I. INTRODUCTION of the pixels of the cover image [6]. Shown in Fig. 2 is the four bits per byte replacement approach of this algorithm wherein Steganography is the art and science of hiding the existence four least significant bits of each color channel of the RGB of communication [1]. Steganography provides secrecy of component is being replaced by the data of the secret message. information by embedding its data into a cover in a form of Because 24-bit pixel data are usually represented in RGB color digital picture, video or audio file [2], in such a way that the format, RGB has been the most commonly used referenced cover file can be openly viewed, played or listened by anybody color space in image steganography [7]. without giving its audience an obvious hint that an underlying information has been hidden in it. Generally, steganography One improvement over the simple LSB Replacement allows extraction of the secret information back unaltered, but algorithm is the Adaptive LSB Embedding algorithm which various techniques have been developed that produces outputs follows a directional embedding technique to achieve with varying aspects of capacity, imperceptibility and robustness [3, 4]. Fig. 2. The bits of the secret message (shaded gray) embedded on each RGB components of the cover image pixels The concept of image steganography is that its cover is in a form of an image file. When the data of the secret message is embedded with the data of the cover image, the resulting image is called the stego image. A very effective image steganography process will produce a stego image that closely resembles and at times visually indistinguishable from the original cover image. By applying a corresponding de- steganography process, the secret message can be extracted back into its original form from the stego image. Shown in Fig. 1 are the inputs and outputs of the steganography and de- steganography process using images for both cover and secret files. 978–1–5386–5538–2/18/$31.00 ©2018 IEEE

2018 15th International Joint Conference on Computer Science and Software Engineering (JCSSE) TABLE II. EXISTING PSNR TEST RESULTS Algorithm LSB Adaptive LSB Replacement Embedding Replaced per byte 3 bits 4 bits 3 bits 4 bits Lena 39.6817 33.6761 40.1097 34.0499 Fig. 3. The bits of the secret message (shaded gray) and the direction bit Mandrill 39.6722 33.6666 40.0957 34.0833 (shaded black) embedded on the pixels of the cover image pixels Pepper 39.7281 33.6510 40.1571 34.0641 maximum image quality in the stego image [8]. This method selects a suitable direction for embedding the byte sections of Average PSNR 39.6940 33.6646 40.1208 34.0658 the secret message that will result the most minimal bit changes between the cover image and the stego image. Shown in Fig. 3 PSNR in the other hand is an engineering term for the is the four bits per byte replacement approach of this algorithm proportion of the maximum possible power of cover image to with the direction status of each byte of the secret message are the power of the differences between cover image and stego stored in the stego image as direction bits. image. A higher is the PSNR value means that there is less distortion in the resulting stego image. Just like in Table I, The image quality of the stego image generated by both the Table II validates the superior quality of stego images produced LSB Replacement Algorithm and the Adaptive LSB on lower bits per byte replacement and with the Adaptive LSB Embedding Algorithm were already tested on various 360x360 Embedding algorithm. 24-bit cover images and random alphanumeric secret message using MSE (Mean Square Error) and PSNR (Peak Signal to III. PROPOSED APPROACH Noise Ratio) [3]. Narrowing down the results of the existing tests to only three of the images shown in Fig. 4, Table I and Not all the aspects of LSB embedding technique can Table II exhibits the significant difference of image quality together be achieved in a high degree in the same stego image. between the three and four bits per byte replacement approach There is always a tradeoff wherein if either capacity, of the two algorithms. imperceptibility, or robustness is to be highly expected from the output, then at least one of the other aspects is given lesser MSE is the average of the difference between the cover priority or achieved in a lesser degree. image and the stego image. The lesser is the value of MSE, the lesser is the distortion of the resulting stego image. Table I The approach proposed in this paper follows the classical shows that three bits per byte replacement preserves the quality image steganography method wherein the LSB side of the of the stego image better compared to its four bits counterpart. RGB bytes of the cover image pixel are the target for The table also proves that the Adaptive LSB Embedding replacement. The method of embedding takes priority on both algorithm produces better result than the LSB Replacement the capacity and imperceptibility aspects of steganography. algorithm. To have a decent secret message capacity of 1/3 the size of the cover image, each byte of the secret message will be embedded over three bytes of the cover image RGB components. Along with that, one bit of each RGB bytes of the cover image will be used to indicate rotate position, and one bit of the blue byte will be used to indicate inversion status. Shown in Fig. 5 is the assignment of bits to create the RGB component of the stego image. (a) Lena (b) Mandrill (c) Pepper Fig. 4. 360x360 Test Images TABLE I. EXISTING MSE TEST RESULTS Algorithm LSB Adaptive LSB Replacement Embedding Replaced per byte 3 bits 4 bits 3 bits 4 bits Lena 7.1174 28.4277 6.3687 25.6978 Mandrill 7.1425 28.4537 6.3928 25.2006 Pepper 7.0271 28.4142 6.2842 25.5579 Fig. 5. Embedding of the cover image, the secret message and other indicator 7.0957 28.4319 6.3486 25.4854 bits into the stego image Average MSE

2018 15th International Joint Conference on Computer Science and Software Engineering (JCSSE) TABLE III. PROPOSED APPROACH AVERAGE MSE AND PSNR RESULTS Cover Replaced per byte: 4 bits Image MSE PSNR Lena 7.4640 39.4011 Mandrill 7.2217 39.5444 Pepper 7.4797 39.3920 Average 7.3885 39.4458 Fig. 6. The rotation sequence of the secret message byte. This process is repeated all throughout the succeeding pixels of the cover image until it embeds all data of the secret message into the stego image. The de-steganography process which extracts back the secret message is done by reading the saved rotate position and inversion status from each pixel of the stego image and then reversely do the embedding process. Though this security measure is not very robust, the added complexity to execute this proposed approach makes it more secure compared to the existing simple and adaptive LSB substitution techniques. Fig. 7. Computation of difference between the rotated replacement bits and its IV. RESULTS AND DISCUSSION embedded indicators, and the corresponding RGB target bits of the cover The quality of the stego image produced by the proposed image. approach is tested using the same image as shown in Fig. 4 as cover images. As for the secret message, an array of randomly Prior embedding, the bits of each byte of the secret message generated data is used. Shown in Table III are the image will be rotated eight times in a sequence as shown in Fig. 6, quality measurement results of the proposed approach using along with the indicator bits that signifies current rotation MSE and PSNR. position and inversion status. The byte rotation generates eight Take note that the proposed approach is a fixed four bits different combinations of the secret message byte as candidates per byte replacement approach. Comparing the average MSE of replacement to the targeted least significant bits of the cover and PSNR results of the proposed approach to the existing tests image. For each rotation, the 3-bit value of the rotation position results of the LSB Replacement and Adaptive LSB Embedding and its 1-bit inversion status is merged with the secret message algorithm as shown in Fig. 8 and 9, the image quality of its byte. Then, the differences of each of the combination’s 4-bit stego image falls between the three and four bits per byte components and its corresponding lower 4-bit RGB byte from replacement approach of the other two LSB algorithms with the cover image is computed as shown in Fig 7. The total values nearer to the three bits per byte approaches. This difference of all color channel will then serve as the difference comparison shows that the proposed approach does replaces score of the combination. The lower is the difference score, the four bits per byte of the cover image but however can produce lesser is the distortion effect of the combination when it stego images with superior visual quality comparative to the replaces the target bits of the cover image. three bits per byte replacement approach of the other algorithm. After its 8th rotation, all bits of the secret message byte are Fig. 8. Average MSE comparison between the 3-bit and 4-bit category of inverted, and then rotated and scored again eight more times. Simple and Adaptive LSB Substitution, and the proposed approach. The inversion will produce new byte value of the secret message and the 2nd eight rotations will generate eight more new combinations in an attempt to find other candidates that has even have lower difference score. Out of the sixteen candidates, the one that has its combination that produced the lowest difference score will have its rotated value, rotate position, and inversion status recalled and then embedded into the stego image in the format previously shown in Fig. 5 in a fixed four bits per byte replacement approach. Because of the numerous candidates generated for embedding selection, the probability of finding and selecting the least distorting combination of the secret message byte is highly increased, and therefore effectively minimizing the distortion of the stego image.

2018 15th International Joint Conference on Computer Science and Software Engineering (JCSSE) Fig. 9. Average PSNR comparison between the 3-bit and 4-bit category of actually help minimizing imperceptibility deterioration by Simple and Adaptive LSB Substitution, and the proposed approach. producing a rotated and/or inverted version of the embedded data in a manner that majority of the pixels of the stego image have values very near if not perfectly equal to the pixels of the cover image. Replacing data bits of an image with equal or near values does a very little or no deterioration to its visual quality, therefore maximizing imperceptibility. Furthermore, the added complexity of the steganography process used by the proposed approach as compared to Simple and Adaptive LSB methods supplements a layer of security in the encryption of the secret message. But without the implementation of security mechanisms such as the use of substitutions, transpositions, or secret keys, its level of robustness is not far behind its other LSB counterparts. It is therefore recommended to conduct future study on this approach with security on the priority. ACKNOWLEDGMENT R. Subong would like to thank Commission of Higher Education of the Philippines and Western Institute of Technology for financially supporting him on his doctoral studies in Technological Institute of the Philippines. The approach in this paper could have not been conceived without their help. REFERENCES Fig. 10. Average selection count of each difference scores. [1] M. S. Sutaone and M. V. Khandare, “Image Based Steganography Using LSB Insertion Technique,” IET Conference on Wireless, Mobile and The difference scores used in this proposed approach have Multimedia Networks, pp. 146 – 151, January 2008. possible values ranging from 0 to 45. Shown in Fig. 10 are the average number of times that each difference score was [2] G.G. Rajput and R. Chavan “A Novel Approach for Image selected on the tested three 360x360 or 129,600-pixels cover Steganography based on LSB Technique,” Proceedings of the images. The graph shows that majority of the selected International Conference on Compute and Data Analysis, pp. 167-170, combinations have difference scores of between 4 to 6. The May 2017 difference score of 7 and below, which modifies only a very little value to the RGB byte of the cover image byte when [3] N. Akhtar, P. Johri and S. Khan, “Enhancing the Security and Quality of embedded, consists a significant 75% share of the LSB based Image Steganography,” 2013 5th International Conference combinations. No difference score with a significant distorting and Computational Intelligence and Communication Networks, pp. 385- value of more than 20 was ever selected in any of the tests. 390, November 2013. This means that the scoring scheme of this proposed approach is very effective in minimizing the difference value between [4] A.A. Jabbar Altaay, S.B. Sahib, and M. Zamani, “An Introduction to the RGB bytes of the cover image and the stego image. This is Image Steganography Techniques,” 2012 International Conference on why it produces less distorted stego image compared to other Advanced Computer Science Applications and Technologies, pp. 122- four bits per byte replacement algorithms. 126, November 2012 V. CONCLUSION AND RECOMMENDATION [5] C.-K. Chan and L.-M. Cheng, “Hiding Data in Images by Simple LSB Substitution,” Pattern Recognition, vol. 37, no. 3, pp. 469–474, March Logically, the more bits of the cover image are being 2004. substituted for other foreign data, the more it is being distorted into an image with lesser imperceptibility. Though this [6] D. Neeta, K. Snehal and D. Jacobs, “Implementation of LSB proposed approach is replacing several bits of the cover image Steganography and Its Evaluation for Various Bits,” 1st International to function as rotate and inversion indicators, those bits Conference on Digital Information Management, pp. 173-178, June 2007. [7] Ö. Çataltaş and K. Tütüncü, “Comparison of LSB Image Steganography Technique in Different Color Spaces,” 2017 International Artificial Intelligence and Data Processing Symposium, pp. 1-6, September 2017. [8] S. Sugathan, “An Improved LSB Embedding Technique for Image Steganography,” 2016 2nd International Conference on Applied and Theoretical Computing and Communication Technology, pp. 609-612, July 2016.