ISSN:43-6999 Journal of Inmaton Communcaton and Intellgence Systems (JICIS) Volume Issue February 06 Steganography System usng Slantlet Transm Ryadh Bassl Abduljabbar Abstract An approach hdng nmaton has been proposed securng nmaton usng Slanlet transm and the T-codes. Same as the wavelet transm the Slantlet transm s better n compresson sgnal and good tme localzaton sgnal compresson than the conventonal transms lke (DCT) dscrete cosne transms. The proposed method provdes effcent securty because the orgnal secret mage s encrypted bee embeddng n order to buld a robust system that s no attacker can defeat t. Some of the well known fdelty measures lke (PSNR and AR) were used to measure the qualty of the Steganography mage and the mage after extracted. The results show that the stego-mage s closed related to the cover mage wth (PSNR) Peak Sgnal to Nose Rato s about 55dB. The recovered secret mage s extracted (00%) f stego-mage has no attack. These methods can provde good hdng capacty and mage qualty. Several types of attacks have been appled to the proposed methods n order to measure the robustness lke (compresson add nose and croppng). The proposed algorthm has been mplemented by usng computer smulaton program MATLAB verson 7.9 under wndows 7 operatng system by Mcrosoft cooperaton. Index Terms dscrete wavelet transm Accuracy Rato Hgh-Hgh band and Hgh-Low band. I. INTRODUCTION Steganography can be consdered as good technque securng nmaton by hdng t n cover mage. Inmaton to be hdden n the cover mage s known as the message and the modfed sgnal whch carres ths message s known as stego. In order to carry the stego data text mage audo and the vdeo are consdered as good meda carryng the hdden data []. Image can be consdered as a wde medum that can be used Steganography. The Steganography of mage depend on the lmted abltes of the Human Vsual System (HVS) all Steganography technques that are used n mage make advantage of the lmted abltes n human eye []. Ryadh Bassl Abduljabbar Dept. of Computer Technques Engneerng Djlah Unversty College (e-mal: ryadh.bassl@duc.edu.q). Baghdad Iraq. There are two wde felds hdng nmaton: Transm doman and the Spatal doman. In the spatal doman mage pxels are processed drectly lke the least sgnfcant bt algorthm. The computatons the spatal doman technques are smple and fast. But the dsadvantage of spatal doman technque s low capablty to bear some sgnal processng or nose. In contrast the transm doman technque the data mage s frst transmed to the frequency doman usng lnear transmatons lke DCT FFT or DWT by hdng the mage data n the coeffcents to get the orgnal mage nverse transm must be permed. In ths way the resulted modfed mage s called stego-mage. For that the frequency doman technques are found to be more robust than the tme doman technques [3]. II. THEORY Transmaton dea can be consdered mostly famlar to mathematcans and engneers. It s powerful tool that s used solvng problems n many felds of scence and engneerng areas nvolvng computer magng. The man dea of transm s changng a mathematcal quantty from one m to another whch t may seem unfamlar but can exhbt good features. Consderng the dscrete wavelet transm (DWT) that mplement by usng flter bank teraton; however but a lmted number of zero moments ths does not produce a dscrete-tme bass whch s very optmal as compared to tme localzaton [4]. The bass of Slantlet transm bass does not depend on flter bank teraton method; nstead of a dversty of flters s used every scale. By consderng two scale DWT terated usng flter bank shown n fgure (-a) and the equvalent m fgure (-b). The flter bank of Slantlet depends on the structure of ts equvalent m shown n fgure (-b) but s occuped by dfferent flters that are not products. For example a two channel case Daubeches flter s the shortest flter that makes the flter bank orthogonal wth K zero moments. For K= zero moments the terated flters of fgure (-b) are of values 0 and 4 but wth the Slantlet flter bank wth K= zero moments shown n fgure () has flter length 8 and 4. Thus the two-scale Slantlet flter bank has a flter length whch s two samples less than that of a two-scale terated Daubeches- flter bank. All Rghts Reserved 06 JICIS
ISSN: 43-6999 Journal of Inmaton Communcaton and Intellgence Systems (JICIS) Volume Issue February 06 Fg. (a) Flter banks DWT (b) The equvalent after usng DWT Manly flters that construct Slantlet bank flters are g (n) f (n) and h (n). The scale length the flters s ί that wll be proportonal to. Ths approxmaton s true nternatonal flter banks; however t s the same as Slantlet flter banks. In general the support of g (n) f (n) and h (n) wll be +. In fgure () t s obvous that the flterbank teraton s has more support than the Slantlet flters. One mportant note s the dfference the general l-scale case. The terated flter bank wth Daubeches length-4 flters analyzes scale wth a flter of length (3. ί )-. On the other hand the Slantlet flter bank analyzes scale wth the flter g (n) of length ί+. Gettng a reducton of samples by ί - the ί scale whch provde good tme localzaton propertes yeld good edges representaton. The mplementaton of Slantlet flter bank wth a parallel structure by usng dfferent flter each scale. The DWT s mplemented n a m of an terated flter bank utlzng a tree structure. By provdng a scale dlaton factor of the Slantlet has less frequency selectve because of the shorter support of component flters whereas DWT flters approxmately provde dlaton factor scale of [7 8]. The flters shown on the rght-hand sde of fgure () are [0 ]: 0 3 0 3 0 0 ( ) G z z z z3 0 4 0 4 0 4 0 4 7 5 55 5 55 9 5 55 7 5 3 55 ( ) F z z z z3 7 5 3 55 4 9 5 55 5 5 55 6 7 5 3 55 z z z z7 3 5 7 ( ) H z z z z3 7 4 5 5 3 6 z z z z 7 III. SLANTLET FILTER BANK PROPERTIES [9-0] Fg. The dfference between two-scale terated D flterbank (left hand sde) and two scale (rght hand sde) Slantlet flterbank. In general the algorthm to obtan the l the scale length flter banks of Slantlet as follows: Consderng l flter banks l-scale Slantlet. Low pass flter (LPF) s the frst flter bank and the transfer functon of t s h (n). The transfer functon of the adjacent to the LPF flter bank has s f (n). Were Both f(n) and h (n) are followed wth a down samplng of l. Consderng the remaned l - flter banks wth transfer functon g (n) and ts shfted tme reverse g( + -)-n) = l - l -. Each g (n) and ts shfted tme reverse must be followed by + down samplng of each value [6].. Slantlet flter bank (equvalently dscrete-tme bass) s orthogonal.. Two zero moments each Slantlet flter bank. 3. Every flter bank has characterstc of an octave-band. 4. Each flter bank has scale-dlaton factor of. 5. The decomposton of each flter bank provdes mult resoluton. The hgh-pass channels are dscarded and passng only the low-pass channel outputs through the synthess flter bank n ths way lower resoluton verson of the orgnal sgnal s obtaned. 6. DWT flter bank dffer from Slantlet flter bank because Slantlet flter s less frequency selectve due to the shorter length of the flters. An mprovement tme localzaton wth a degradaton of frequency selectvty. 7. The flters of Slantlet transm are consdered to be pecewse lnear. 8. Fast because no structure tree and mplemented parallel structure. 9. Gves a hgh order of approxmaton. All Rghts Reserved 06 JICIS
ISSN:43-6999 Journal of Inmaton Communcaton and Intellgence Systems (JICIS) Volume Issue February 06 IV. DERIVATION OF THE SLANTLET FILTERS COEFFICIENTS Because the sought-after flter g (n) s to be lnear t s descrbed by four parameters and can be wrtten as: a g ( n) a 00 0 a a 0 n ( n ) n 0... n... For each condton an algebrac equaton can be wrtten as n terms of the four parameters a 00 a 0 a 0 and a to obtan g (n) Where: t 3 (.( )) m m 0 s.( m ) 0 (( m ). s 3 mt)( m ) m 00 ( s0 t0) 0 ( s0 t0) a0 ( s t ) m s t a a a ( s ) t Note that the parameters a 00 a 0 a 0 and a depend on The same approach works f (n) and h (n). Usng agan a pecewse lnear m f (n) and h (n) can be wrtten n terms of eght unknown parameters b 00 b 0 b 0 b c 00 c 0 c 0 and c [9] The hgher symbols frequency occurrence codes are shorter than those codes the symbols that have a lower frequency occurrence. The Huffman technque depends on the creaton of nodes of bnary tree m. Type s nodes are ether a leaf code or an nternal node. At the begnnng all nodes are leaf nodes leaf codes have a symbol and weght (number of appearance) of the symbol and optonally a lnk to a parent node whch makes t easy to read the code (n reverse) startng from a leaf node. The Internal nodes contan symbol weght lnks to two small nodes an optonal lnk to a base node. Where bt zero represents followng the left chld and bt one represents the rght chld. A complete tree has n- nternal nodes and n leaf nodes. The man operaton starts wth the leaf nodes wth the probabltes of the represented symbols they contan The creaton of the chldren nodes wth smallest probablty such that the probablty of the new node s s equal to the probablty of sum of the chldren. Then the prevous nodes are merged nto one node and wth the new node beng now consdered the procedure s terated untl one node remans [-]. The Proposed procedure The procedure s mplemented usng the proposed Steganography system s used non-blnd method; t s called non-blnd because t needs the exact mage to recover the data (mage or text). In ths proposed method embeds ether text or mage n cover mage of gray scale or color. Hdng Image n Image Many stages are needed n order to hde mage n mage based on non-blnd method as n fgure (3). b h ( n) b 00 0 c f ( n) c 00 0 b b c c 0 n ( n ) 0 n ( n ) n 0... n... n 0... n... Huffman Encodng The mappng of one code word to one symbol s optmal wth Huffman codes. The secret text got each letter a bnary code assgned by Huffman; by usng mathematcal propertes of the source stream characters and the producton of the respectve codes of characters s done. The produced codes are of varable length an ntegral number of bts s used. Fg. 3 Block dagram hdng usng proposed method Encrypton Process: The secret mage frst encrypted by combnatonal permutaton encrypton Block Pxel and Bt permutaton as descrbed n the frst method. Converson Process: In ths stage the encrypted secret mage converted to bnary stream. Each pxel All Rghts Reserved 06 JICIS 3
ISSN: 43-6999 Journal of Inmaton Communcaton and Intellgence Systems (JICIS) Volume Issue February 06 converts from decmal to bnary to obtan D array of bnary then reshape D to D array of bnary stream. Embeddng Process: In ths process after the converson stage s completed frstly the cover mage s taken and -D (DSLT) s appled to the mage the second step s taken converted secret mage and embeddng one bt of bnary per pxel n cover mage. Durng embeddng the sze of the secret mage should be smaller enough than the cover mage. The embeddng steps are shown n fgure below and algorthm. k q: Scalng factor cover and secret mage. Step8: Obtan the encrypted-mage by usng the nverse of DSLT Step9: Fnsh. When the orgnal mage and carrer mage are both color the steps of embeddng and extracton process each color plane of mage are permed separately then combne the resultant of the three color planes n order to obtan color stego-mage. Extracton Process: When the recever wants to extract the secret mage many steps appled n the embeddng process are appled n a reverse manner as explaned below n algorthm 4 and fgure (5). Algorthm : Extracton Steps Second Proposed Method Input: Stego-mage Scalng factor (kq) Orgnal mage Keys Output: Secret mage. Step: The dmenson of encrypted mage should be power of. Step: Use Slantlet method to stego-mage and orgnal mage to get 4-subbands (LL LH HL and HH). Fg. 4 Schematc of embeddng steps usng proposed method Algorthm : Embeddng Steps Second Proposed Method Input: Secret mage Cover mage Scalng factor (q k). Output: Stego-mage. Step: The dmenson of carrer mage should be power of f not by requred paddng. Step: Encrypt the orgnal mage. Step3: Transm the encrypted orgnal mage to D bnary sequence. Step4: Convert the D bnary sequence to one dmenson bnary sequence. Step3: Extract the bnary strng by applyng the reverse of alpha blendng as n equaton: Sc=(S- k * C)/q Step4: Convert the bnary strng to decmal. Step5: Reshape from D decmal to D decmal mage. Step6: Reconstruct the encrypted secret mage by nverse DSLT. Step7: Apply decrypton steps based on keys. Step8: Dsplay secret mage. Step9: End. Step5: Apply Slantlet transm to the orgnal mage to get (LL LH HL and HH) sub bands by usng -D SLT. Step6: Check over flow. Step7: Embed the bnary sequence accordng to the mula of alpha blendng as n equaton: S=k*C + q*sc Where C: The coeffcent of cover mage Sc: Bnary stream of secret mage. All Rghts Reserved 06 JICIS 4
ISSN:43-6999 Journal of Inmaton Communcaton and Intellgence Systems (JICIS) Volume Issue February 06 k q: scalng factor cover mage and secret text where each of k and q equal 0.000. Step6: Reconstruct the stego-mage by applyng IDSLT. Step6: End. Fg. 5 Schematc of extractng steps usng proposed method Hdng Text n Image: By usng the same embeddng method alpha blendng method secret text can be embedded n gray or color mage and the only dfference s that the text s encoded and compressed by Huffman codng rather than permutaton encrypton. A bref descrpton the stages of embeddng s explaned as follow and s shown n fgure (6). Huffman Encodng Secret text: Frst the orgnal text should be encoded usng Huffman codng bee embeddng t nto carrer mage t. Bt stream length must be less than the space needs to embed n cover mage. Embeddng Secret Text n Image: The followng algorthm 5 and fgure (6) descrbe the embeddng steps hdng text n mage: Algorthm 3: Embeddng Secret Text second proposed method Input: Cover mage Text fle Scalng factor (k q) and Huffman table Output: Stego-Image Step: The dmenson of orgnal mage should be power of f not by requred paddng. Step: Apply Slantlet transm to carrer mage to get 4-subbands (LL LH HL and HH) by usng -D SLT. Step3: Encode the orgnal text by Huffman code as bnary sequence. Step4: Check the over flow of cover mage. Step5: Embed the bnary stream nto hgh frequency coeffcent of cover mage accordng to the mula of alpha blendng as n equaton: S=k * C + q * Scr Where: C: coeffcent of cover mage Scr: bnary stream of secret text Fg. 6 Schematc of embeddng text n mage steps usng proposed method Extracton Text from Image: To obtan the orgnal text from the encrypted mage the followng steps descrbed n algorthm 5 are appled and shown n fgure (7). Algorthm 4: Extracton text from mage Input: Stego-mage Scalng factor (k q) Orgnal mage Huffman table Output: Secret text Step: Dmenson of orgnal mage should be power of f not by requred paddng. Step: Use Slantlet transm to the encrypted mage and orgnal mage to get 4-subbands (LL LH HL and HH). Step3: Extract the bnary strng from hgh frequency coeffcent by applyng the reverse of Alpha-blendng as n equaton Scr =(S - k*c)/q Step4: Reconstruct the text from bnary strng by usng Huffman decoder steps. Step5: Dsplay the secret text. Step6: End. All Rghts Reserved 06 JICIS 5
ISSN: 43-6999 Journal of Inmaton Communcaton and Intellgence Systems (JICIS) Volume Issue February 06 Theree the TCP source can estmate the bottleneck Fg. 7 Schematc of extracton text from mage steps usng proposed method V. RESULTS Usng the weght method fgure (8) and fgure (9) show the results embeddng and extractng steps where the test mages used are gray mages (Barbara 5*5 and Ensten 8 *8). Fg. 9 The extracton steps weght method usng gray mages In fgure (0) the result of hstogram carrer mage hdden mage and the dfference between the two hstograms are shown. A sental data amount per RTT perod as shown n Eq. 3: Fg. 0 Hstogram embeddng and extractng gray mage usng second proposed method REFERENCES [] Faben A. P. Pettcolas Ross J. Anderson and Markus G. Kuhn Inmaton Hdng A Survey Proceedngs of the IEEE specal ssue on protecton of multmeda content Vol 87 No 7 pp.06-078 July 999. [] Al-Saedy M. "Image n Image Steganography Usng Fourer Transm Codng" M.Sc. thess n computer scence Unversty of Technology 004. [3] Ndaa Hassan Abbas Robust and Hgh Capacty Watermarkng System usng Error Correctng Codes M.Sc.thess Electrcal Eng. Dep. Al-Mustansryah Unversty 005. [4] Mutt S.K. and Sushl Kumar Secure Image Steganography Based on Slantlet Transm Internatonal Conference on Methods and Models n Computer Scence 009. [5] Mutt S.K. and Sushl Kumar Data Hdng Technques Based On Wavelet-Lke Transm and Complex Wavelet Transms IEEE 00. [6] Al Daneshkhah Hassan Aghaena Seyed Hamed Seyed A More Secure Steganography Method n Spatal Doman Second Internatonal Conference on Intellgent Systems Modellng and Smulaton 0. [7] Khzar Hayat Wavelet Based Data Hdng of Dem n the Context of Realtme 3D Vsualzaton M.Sc.thess Master Research n Inmatcs Unversty de Montpeller II 005-06. [8] Navneet Kumar Mandhan Watermarkng usng Decmal Sequence A Thess Lousana State Unversty and Agrcultural and Mechancal College n partal fulfllment of the requrements the degree of Master of Scence n Electrcal Engneerng August 004. [9] Katzenbesser S. Pettcolas F.A.P. Inmaton Hdng Technques Steganography and Dgtal Watermarkng Artech House Publshers 000. [0] Al Al-Ataby and Fawz Al-Nama A Modfed Hgh Capacty Image Steganography Technque Based on Wavelet Transm The Internatonal Arab Journal of Inmaton Technology Vol. 7 No. 4 October 00. [] Hkmat N. Abdullah Safa'a A. Al "Implementaton of 8-Pont Slantlet Transm Based Polynomal Cancellaton Codng-OFDM System Usng FPGA" 7th Internatonal Mult-Conference on Systems Sgnals and Devces 00. [] Cormen T.H. Leserson C.E. Rvest R.L. & Sten C. Introducton to Algorthms Second Edton. MIT Press and McGraw-Hll 00. Fg. 0 Hstogram embeddng and extractng gray mage usng second proposed method All Rghts Reserved 06 JICIS 6