Internatonal Journal of Computer Technology and Electroncs Engneerng (IJCTEE) Volume 3 Issue 5 October 2013 A Novel and Secure technque for mage Abstract: Steganography s the art and scence of hdng nformaton n dgtal cover meda n a manner that the exstence of the message s unknown..it s an applcaton under nformaton securty feld.. The am of ths paper s to propose a novel and secure hgh-capacty mage steganography technque that depends on wavelet transform wth acceptable levels of mperceptblty and dstorton n the cover mage and hgh level of overall securty.our method works as follows:frst secrete nformaton s hdden nsde the grey scale mage whle mantanng good vsual qualty and hgh capacty of secret nformaton and then encrypts the mage for securty purpose. Advantage of ths method s full reconstructon of secret data wthout havng orgnal mage present on the recpent sde. Accordng to the smulaton results the proposed approach provdes fne mage qualty and enough embeddng capacty. Furthermore respectable securty s mantaned as well snce no message can be extracted wthout the Key matrx and the decodng rules. Keywords:Steganographynformaton securtydwt I.INTRODUCTION Watermarkng data hdng and steganography are three very closely related felds that share many techncal approaches and possble to have a great deal of overlap. However there are fundamental dfferences between them that affect the overall requrements and thus the desgn of a techncal soluton [1]. Dgtal watermarkng s mostly used n copyrght protecton but steganography s a technque of embeddng the secret message nto a dgtal meda n a manner that the exstence of the message s unknown whch ensure that unntended recpents wll not be aware of the exstence of the embedded secret data n cover meda [2]. So whole system can be consdered as secret communcaton. Concept of steganography s dfferent from cryptography Cryptography encrypts a message wth the help of certan cryptographc algorthms for convertng the secret data nto unntellgble form. whereas steganography hdes the message n cover mage so that t becomes nvsble. Unlke cryptography steganography conceals the fact that there s secret communcaton gong on and stll mage may be represented as well suted dgtal meda for embeddng of secrete data. Even more the advantage of cryptography technques are executed on secret data before embeddng nto stll mage; to strengthen Steganography usng DWT Vkas pratap sngh Prof. Shrkant Lade 5 securty level and also to suppress the energy compacton of secret data [3]. Steganalyss s the scence of detectng hdden nformaton nto dgtal meda. breakng steganography s the man purpose of Steganalyss and the detecton of stego mage s the goal of steganalyss. All steganalyss algorthms depend on the Steganographc algorthms that ntroduce statstcal dfferences between cover and stego mage. Steganalyss deals wth three mportant categores: Vsual attacks: In these types of attacks wth a assstance of a computer or through nspecton wth a naked eye t reveal the presence of hdden nformaton whch helps to separate the mage nto bt planes for further more analyss. Statstcal attacks: These types of attacks are more powerful and successful because they reveal the smallest alteratons n an mages statstcal behavor. Statstcal attacks can be further dvded nto Passve attack and Actve attack. Passve attacks nvolves wth dentfyng presence or absence of a covert message or the embeddng algorthm used etc. whereas the goal of actve attacks s to estmate the embedded message length or the locatons of the hdden message or the secret key used n embeddng. Thrdly structural attacks are based on fact that format of the data fles often changes as the data to be hdden are embedded; on dentfyng these characterstc structure changes can detect the exstence of mage [4]. Research n hdng nformaton nsde mage usng steganography technque has been developed by many researchers for example n [5-10]. Warkentn et al. [5] proposed an method to hde data nsde the audovsual fles. In ther algorthm to hde data the secret message has to be hdden nto a cover mage. El-Emam [6] on the other hand proposed steganography algorthms that hde a large amount of data wth hgh securty. hs steganography algorthm s based on hdng a large amount of data (mage audo text) fle nsde a color btmap (bmp) mage. In hs research the mage wll be fltered and segmented where bts replacement s used on the approprate pxels. These pxels are selected randomly rather than sequentally. Wu et al. [7] used pxel-value dfferencng by parttonng the orgnal mage nto non-overlappng blocks of two consecutve pxels.and hde the message n these blocks. Lee and Chen s method [8] proposed a method n whch the least sgnfcant bt (LSB) of each pxel n the cover mage s modfed to embed a secret message. In Wang et al. s method [9] the optmal substtuton of LSB s exploted.
Internatonal Journal of Computer Technology and Electroncs Engneerng (IJCTEE) Volume 3 Issue 5 October 2013 These schemes have a hgher qualty stego-mage but are senstve to modfcaton. Many practcal tests propose to use the Wavelet transform doman for steganography because of a number of advantages that can be ganed by usng ths approach. The use of such transform wll manly address the capacty and robustness of the Informaton- Hdng system features. The work descrbed n ths paper mplements steganography n the Wavelet doman. The herarchcal nature of the Wavelet representaton allows mult-resolutonal detecton of the hdden message whch s a Gaussan dstrbuted random vector added to all the hgh pass bands n the Wavelet doman. It s shown that when subjected to dstorton from compresson the correspondng hdden message can stll be correctly dentfed at each resoluton n the Dscrete Wavelet Transform (DWT) doman [110 11 12]. In ths paper we propose a novel and secure method utlzng the features obtaned from DWT coeffcents to hde secret nformaton n grey scale mage whle mantanng good vsual qualty and hgh capacty of secret nformaton and then encypted the mage for securty purpose. Advantage of ths method s full reconstructon of secret data wthout havng orgnal mage present on the recpent sde. The organzaton of the paper s as follows.. In the next secton we gve overvew of proposed method based on DWT whch coeffcents were used for embeddng wth specfc way. next we descrbe objectve vsual qualty measurements to smulate human percepton whch are PSNR[dB] and MSE modfed by Contrast Senstvty Functon to better reflecton of Human Vsual System (HVS) model. In Secton 5 are gven results of proposed method n manner of PSNR value usng cover mage wth dfferent detal levels.. Fnally we conclude our paper n Secton 6 wth dscusson and contrbuton of our proposed method to the mage steganography. II.PROPOSED IMAGE STEGNAGRAPHY METHOD In ths secton we propose a new frequency doman method for mage steganography. The mert s to ncrease mage qualty by hdng the messages n HL LH and HH sub-bands whle keepng LL sub-band nvarant. The detaled descrpton for both embeddng and extracton s provded n the followng two subsectons step by step. Suppose X s the orgnal 8-bt gray-level cover-mage of M C N C pxels. It s denoted as: X { x 1 M 1 j N x {01...255}}...(1) j C C j let P s the n-bt secret message represented as: P { p 1 n p {01}}...(2) Next use the random number to generate the prvate key of length same as the length of message because the sze of encrypt message s equal to the orgnal message then apply X-OR operator to generate the cpher message of length n bts. Where S P K K { k 1 n k {01}}...(3) and S { s 1 n c {01} and s p k }...(4) A.2 DWT Decomposton Apply DWT on X to obtan the frequency doman matrx H. The 4 sub-bands obtaned are denoted as H LL H HL H LH and H HH (All 4 sub-bands have the same sze of M C /2 N C /2). The coeffcents sutable for embeddng that are elected from detal areas (H HL H LH and H HH ) are also denoted as A. Embeddng procedure A.1 Input secret message and cover mage C { c 1 M 1 j N }...(5) j C C And secret message as S { s 1 n c {01}} that s reszed to same dmenson as C. 6
Internatonal Journal of Computer Technology and Electroncs Engneerng (IJCTEE) Volume 3 Issue 5 October 2013 A.3 Message Embeddng Stage B. Extractng Algorthm The process of embeddng conssts of comparng the secret message S j wth detal coeffcents C j n form of followng sequence: f S íj == 0 ^ C j mod 1 0 then C j = C j f S íj == 0 ^ C j mod 1 = 0 then C j = sgn(c j )( C j 1/2) f S íj == 1 ^ C j mod 1 0 then C j = sgn(c j )(C j rem(c j )) f S íj == 1^ C j mod 1 = 0 then C j = C j where C j are modfed coeffcents of DWT detal area. A.4 Stego-Sgnal Reconstructon Stage Ths stage all the modfed detaled coffecent are converted back from frequency doman to tme doman. The IDWT s used to reconstruct the segments of stego- mage based on modfed Detaled coeffcent and unmodfed Approxmate coeffcent. After embeddng all secret data and performng nverse DWT (IDWT) on C j the stego mage E s obtaned. A.5 Block Level XOR Operaton Use the stego mage E as the nput for block level XOR operaton and encrypt t usng 4.2 to generate the fnal encrypted mage Y.. The stego mage E s dvded nto 2 pxels 2 pxels blocks. The mage pxel contaned by every block B j of E s encrypted usng block level XOR operaton by four eght bt sub key K 1 K 2 K 3 and K 4. as gven below Respectvely. 1.1 1.1 1...(6) 1.2 1.2 2 2.1 2.1 3 2.2 2.2 4 Where P j s the pxel value at th and jth locaton n block nsde pxel of mage.the encrypted mage by usng the XOR operaton s called by cpher mage Y and t s ready to be sent to recever ste. The total sze of key n our algorthm s 32 bt long whch proves to be strong enough.. The extracton process s dvded nto blocks stego segmentaton wave decomposton Coeffcent selecton Message extracton and reverse encrypton. The Quantzaton parameter Q needs to be the same that s used durng encrypton. B.1 Input Encrypted Stego Image In the message recovery algorthm frst we select the stego mage Y from whch data s to be extracted. Let the 8-bt graylevel stego-mage of MC NC pxels s represented as: Y { y 1 M 1 j N y {01...255}}...(7) j C C j B.2 Block Level X-OR Operaton Next Step s to performs the block level X-OR operaton on encrypted stego mage Y by usng the X-OR operaton. Agan we dvded Cpher mage Y nto 2 pxels 2 pxels blocks. Then mage pxel contaned by every pxel B j of Y s Decrypted usng block level X-OR operaton by Same four eght bt keys (K 1 K 2 K 3 K 4 ) as followed: Decrypton P 1.1 1.1 1 Decrypton P asp K...(8) Decrypton P Decrypton P asp K 1.2 1.2 2 asp K 2.1 2.1 3 asp K 2.2 2.2 4 B.3 DWT Decomposton Next apply DWT on E to obtan the frequency doman matrx H. The 4 sub-bands obtaned are denoted as H LL H LH H L H HH (All 4 sub-bands have the same sze of MC/2 NC/2). Agan Normalzng coeffcents n H HL H LH and H HH subbands as done at the tme of encrypton. B.4 Secret Message Recovery Stage:Secret message recovery stage s very smple and based on modulo operaton of stego mage coeffcents C j n followng way: 7
Internatonal Journal of Computer Technology and Electroncs Engneerng (IJCTEE) Volume 3 Issue 5 October 2013 1 f C' mod 1= 0 '...(9) 0 f C' mod 1 0 S$ j j j B.5 Reverse Encrypton Before delvery of the secret message S to recever t must be converted back to t s orgnal form.to do so frst we use the same random number to generate the prvate key of length same as the length of message then apply X-OR operator to get the orgnal message of length N bts. next we convert message of t s bnary form to text format.and then delver to the recever. C. Qualty Measurement The qualty of the stego mage and the extracted secret mage s measured by calculaton of certan qualty measurement metrcs [13]. These metrcs gves the comparson rato between the orgnal mage and the modfed mage. The qualty may be assessed on the bass of these values. The metrcs used n ths paper are as follows: 1 MSE f j f j MN M N 2 ( ( ) '( ))...(11) 1 j 1 Where f ( j) s the orgnal mage and f ( j) s the stego mage. A large value for MSE means that the mage s of poor qualty. D. Experment Setup Proposed technque are mplemented on Wndows PC havng Intel 2.4 GHz processor and 2GB RAM and run usng Matlab 9a. We have consdered fve dfferent mage fles n ths experment to embed dgtal data. They are Lena.bmp (27.3 KB) Boat.bmp (71.5 KB) Baboon.bmp (61.2 KB) Papper.bmp (192 KB) and Man.bmp (50.8 KB). all the mages are 8 bt gray scale mages and the dmenson of all the mage s 256 256 pxels shown n Fgure 1. We appled Haar wavelets on cover mage and choose one of the detaled coeffcent where the data s to be hde. C.1 Peak sgnal to nose rato (PSNR) The PSNR depcts the measure of reconstructon of the compressed mage. Ths metrc s used for dscrmnatng between the cover and stego mage [14]. The easy computaton s the advantage of ths measure. It s formulated as: (a) Leena (b) Baboon PSNR 2 10 log 255 / MSE...(10) A low value of PSNR shows that the constructed mage s of poor qualty. (c) Boat (d) Papper C.2 Mean square error (MSE) MSE s one of the most frequently used qualty measurement technque followed by PSNR. The MSE [14] can be defned as the measure of average of the squares of the dfference between the ntenstes of the stego mage and the cover mage. It s popularly used because of the mathematcal tractablty t offers. It s represented as: (e) Man Fgure1: Test Image (256x256 pxels) Used n our Experments III. PERFORMANCE ANALYSIS 8
Internatonal Journal of Computer Technology and Electroncs Engneerng (IJCTEE) Volume 3 Issue 5 October 2013 The tests were performed wth a gray standard testng mage Lena and other cover mage depcted n Fg. 1 wth the same sze 256x256 pxels (contanng 64 kb). The secret message what represented by (contanng 2048B). The PSNR values (qualty) of obtaned stego mages (Table 1). The qualty of our method s stego mages s good and acceptable compared to the stego mages qualty of other three steganography methods. / Image Lena Pepper s Baboo n Grl Arpla ne JSteg 0.4521 0.4525 0.4523 0.4541 0.4548 F5 0.4523 0.4476 0.4437 0.4526 0.4524 In our method we kept a consderable number of coeffcents to represent each block hence stego mages of our method have good qualty. Even though we have slghtly modfed the mddle frequency coeffcents t gves better results than dscardng such coeffcents. However we embedded the secret bts by modfyng low coeffcents; therefore the stego mages of our method have good qualty Table I: The PSNR(dB) of Our proposed Steganography and other method consdered 16x16 Quantza ton Our proposed 0.1854 0.1846 0.1731 0.1831 0.1854 0.5384 0.5284 0.5234 0.5237 0.5245 / Image Lena Peppe rs Baboon Grl Arplane JSteg 27.10 27.50 25.51 28.72 26.36 F5 27.41 28.23 28.13 28.11 27.11 JMQT 31.20 32.77 30.10 32.07 31.03 16x16 Quantzaton Our proposed 30.12 32.17 26.14 32.15 30.17 47.23 47.23 47.20 47.7 46.28 The computatonal tme of the proposed method on dfferent test mages wth other steganography methods (Table 2). The runnng tme of our method s almost double of that n the JSteg and JMQT methods. The steganographc capacty and stego mage mperceptblty represent the two man requrements of mage steganography methods (Zhang and Wang 2005). As a result our steganography method provdes larger steganographc capacty and acceptable stego mage qualty whch makes t superor to the other steganography methods tested. Table II: The Executon tme (sec) of Our proposed Steganography and other method consdered IV. CONCLUSIONS The research am of ths dssertaton s to ncrease the steganographc capacty and mprove the qualty of stego mages In ths paper we proposed a novel and secure mage steganography method utlzng the features obtaned from DWT coeffcents. Qualty of embedded mages s ncreased because the secret messages are embedded n the hgh frequency subbands whch human eyes are less senstve to. Accordng to the smulaton results the proposed approach provdes fne mage qualty and enough embeddng capacty. Furthermore respectable securty s mantaned as well snce no message can be extracted wthout the Key matrx and the decodng rules. REFERENCES [1] Vladmír BÁNOCI Gabrel BUGÁR Dušan LEVICKÝ A Novel of Image Steganography n DWT Doman [2]. COX I. J. MILLER M. L. B KALKER T. Dgtal watermarkng and steganography. 2nd ed. Burlngton: Morgan Kaufmann 200BLOOM J. A. FRIDRICH J. 08. [3]N.N. El-Emam. (2007). Hdng a large amount of data wth hgh securty usngsteganography algorthm. Computer Scence. [On-lne]. 3(4) pp. 223-232. 2011. [4] Souvk Bhattacharyya Gautam Sanyal A Robust Image Steganography usng DWT Dfference Modulaton (DWTDM) J. Computer Network and Informaton Securty 2012 7 27-40. [5]. M. Warkentn M.B. Schmdt E. BekkerngSteganography and steganalyss Premer reference Source Intellectual Property Protecton for Multmeda Informaton technology Chapter XIX 2008 pp. 374-380. [6] N.N. El-Emam Hdng a large amount of data wth hgh securty usng steganography algorthm Journal of Computer Scence 3 (2007) 223-232. 9
Internatonal Journal of Computer Technology and Electroncs Engneerng (IJCTEE) Volume 3 Issue 5 October 2013 [7] P.C. Wu W.H. Tsa A steganographc method for mages by pxel-value dfferencng Pattern Recognton Letters 24 (2003) 1613-1626. [8] LEE Y.K. CHEN L. H. Hgh cap pacty mage model. IEEE Proc. - Vson Image Sgnal Process. 2000 Vol. 147 no.3 p. 288-294. [9] WANG R. Z. LIN C. F. LIN J. C. Image hdng by optmal LSB substtuton and genetc algorthm. Pattern recog. 2001 Vol. 34 no. 3 p. 671-683. [10] Blgn A. Sementll J. Sheng F. and MarcellnW. Scalable Image Codng Usng Reversble Integer Wavelet Transforms Computer Journal of Image Processng IEEE Transactons vol. 9 no. 4 pp. 1972-1977 2000. [11] Tolbal F. Ghonemy A. Taha A. and Khalfa S. Usng Integer Wavelet Transforms n Colored Image Steganography Internatonal Journal on Intellgent Cooperatve Informaton Systems vol. 4 no. 2 pp. 230-235 2004. [12] Walker S. A Premer of Wavelets and Ther Scentfc Applcatons CRC Press 1999. [13] Abbas Cheddad Joan Condell Kevn Curran Paul Mc Kevtt Dgtal mage steganography: Survey and analyss of current methods Sgnal Processng 90 (2010) 727 752. [14] W. Bender D. Gruhl N. Mormoto A. Lu Technques for data hdng IBM Syst. J. 35 (3&4) (1996) 313 336 10