Information Hiding Method Based on Block DWT Sub-Band Feature Encoding

Transcription

1 J. Sofware Engineering & Applicaion, 9, : doi:.436/jea.9.55 Publihed Online December 9 (hp:// 383 Informaion Hiding Mehod Baed on Bloc DWT Sub-Band Feaure Encoding Qiudong SUN, Wenxin MA, Wenying YAN, Hong DAI School of Elecronic and Elecrical Engineering, Shanghai Second Polyechnic Univeriy, Shanghai, China. {qdun, wxma, wyyan, daihong}@ee.pu.cn Received Augu h, 9; revied Sepember 7 h, 9; acceped Ocober h, 9. ABSTRACT For realizing of long informaion hiding and cover communicaion, a binary waermar equence wa obained firly from a file and encoded by a redundan encoding mehod. Then, wo neighboring bloc were eleced a each ime from he Hilber canning equence of carrier image bloc, and ranformed by -level dicree wavele ranformaion (DWT). And hen he double bloc baed JND (ju noiceable difference) were calculaed wih a viual model. According o he differen code of each wo waermar bi, he average value of wo correponding deail ub-band were modified by uing one of JND o hide informaion ino carrier image. The perimenal reul how ha he hidden informaion i inviible o human eye, and he algorihm i robu o ome common image proceing operaion. The concluion i ha he algorihm i effecive and pracical. Keyword: Sub-Band Feaure Encoding, Redundan Encoding, Viual Model, Dicree Wavele Tranformaion, Informaion Hiding. Inroducion Wih he developmen of informaion echnology, people have paid more and more aenion o he informaion ecuriy. Informaion hiding in a digial image ha become he focu of he informaion ecuriy reearch. For an effecive informaion hiding cheme, hree baic requiremen hould be aified: ranparency, robune and ecuriy. The former wo are in conflic wih each oher. To diolve hi conflic availably, we can conider uing he maing characeriic of human viual yem (HVS) []. Duo o i good ime-frequency localizaion funcion i imilar o he viual maing of HVS, he DWT ha been ued widely in he field of informaion hiding []. A good hiding echnique hould alo rac he hidden informaion from ego-image blindly. In recen year, many algorihm baed on HVS and DWT had been propoed for informaion hiding [ 6]. And ome [ 3] of hem alo had implemened he blind racion of hidden informaion. Bu hoe algorihm are moly armed a binary iconic waermar. So hey are unuiable for hiding he informaion and cover communicaion. The reference [5] propoed a robu encrypion echnique for informaion. Though he ranparency, robune and ecuriy of ha algorihm were all good, he embedded capaciy wa rericed due o only one bi waermar can be hidden in wo bloc, whoe ize were eled a 8 8. Ue for reference [5] in waermar embedding, we propoe an adapive informaion hiding mehod baed on average value relaion of correponding DWT ub-band of wo neighboring bloc wih double JND hrehold and adjuable bloc ize. A menioned previouly, in order o adju he inpu image for ranparen waermar, we employ a viual model [,5] o calculae he differen double bloc baed JND hrehold for deermining he ineniy of waermaring a he differen locaion of image. We alo give a redundan encoding mehod for robune. Thi paper i organized a follow. In Secion, we give he JND hrehold calculaion equaion for conrolling he embedding ineniy. Secion 3 preen he informaion hiding algorihm and i racion in deail. Secion 4 amine he performance of propoed algorihm, and how ha he propoed cheme yield more effecive and beer performance, boh in erm of ranparency and robune hrough imulaion. Secion 5 give he concluion of hi paper.. JND Threhold Calculaion. Viual Model Under he bacground gray f, he human eye relaive eniiviy o gray change (f)=f/f, which i a non-linear funcion of f, can be approximaed by he equaion a follow [5]: Copyrigh 9 SciRe

2 384 Informaion Hiding Mehod Baed on Bloc DWT Sub-Band Feaure Encoding 8 f f (56 f ) 8 ( f ).e e () f where e i he bae of naural logarihm. In perimen, we can ue he gray mean of K K image bloc B locaed a (u, v) a he bacground gray f, i.e. f=mean (B ).. JND Calculaion To enure he waermar ha good ranparency and robune, we can ue JND o adju he ineniy of waermar-embedding [,4,5]. The image bloc B i DWT-ranformed ino an approximae image and hree deail ub-band image D ( HH, HL, LH, repreen he hree deail ub-band of diagonal, horizonal and verical direcion repecively). The JND of hree deail ub-band are repreened a follow: J T F () where T i he normalized value of T f f E a he range [a,b], while E i he normalized enropy of B. When HH, F equal, oherwie i i [4]. 3. Adapive Informaion Hiding Scheme 3. Waermar Embedding Le W repreen a waermar equence, B and B are wo neighboring image bloc and heir DWT-ran formed deail ub-band are D and D (imply mared by D and D, or by a univeral ymbol D, {, }). Now we can define he admiible diorion facor of ub-band coefficien of DWT a follow: D λ {,} (3) mean( D ) where i a poiive number, which i an effec facor of abolue value of deail ub-band coefficien o embedding ineniy. When he bloc ize i, whaever he value of i, he Equaion (3) i conan and can be implified o λ. We aume ha i he mean value of JND of wo neighboring bloc. I i repreened by equaion a follow: J J (4) If he cryic normalized range of [a,b] in Equaion () i e by wo differen un-overlapped acend range [a,b ] and [a,b ], uch a [,] and [6,7], hen we can ge wo differen from Equaion () and (4). They can be repreened by and, or by a univeral ymbol r, r {,}. We alo aume ha d i he correponding DWT deail ub-band coefficien difference of wo neighboring bloc a ame direcion, and i he adjumen ineniy marix of deail ub-band coefficien. They are repreened by equaion a follow repecively: W mean( D ) mean( D ) d Sign (5) ε Sign (6) W W λ d where Sign( ) i a ign funcion, i i defined a follow: Sign x x even x odd W i he -h elemen of binary waermar equence W. If he DWT deail ub-band feaure of wo neighboring bloc afer embedded hould aify he relaionhip wih he conecuive wo bi W and W + of waermar equence a hown in Table. We can prove ha he waermar embedding rule i a follow: D ε D D W and d or W (7) if oherwie From he embedding rule, we now ha each couple correponding deail ub-band of wo neighboring bloc can be embedded waermar bi. And he wo neighboring bloc have hree couple correponding deail ub-band. So, if he ize of carrier image i M N, he informaion hiding capaciy of hi algorihm can reach o he value of (3MN)/(K ) bi. I i double han ha of [5]. For ample, if he ize of carrier image i 5 5 and he bloc ize i, hen he full informaion hiding capaciy i 9668 bi or 4576 bye. I i large enough o hide informaion. If applied o hide hor informaion ino a carrier image, hi mehod can bring an enough redundancy o enure i robune. 3. Waermar Exracion Being he ame wih waermar embedding, we hould elec wo neighboring bloc B and B each ime from Hilber canning equence of ego-image bloc, and a couple of heir DWT deail ub-band Dˆ and Dˆ. And e h=(b +a )/. Then, we can prove ha he waermar racion rule i a follow: Table. The relaionhip beween waermar code and he DWT deail ub-band feaure of wo neighboring bloc The ize relaionhip of correponding W and W + DWT deail ub-band mean( D ) mean( D ) mean( D ) mean( D ) mean( D ) mean( D) mean( D ) mean( D ) Copyrigh 9 SciRe

3 Informaion Hiding Mehod Baed on Bloc DWT Sub-Band Feaure Encoding 385 ˆ W,, if mean( Dˆ ) mean( Dˆ ) ele W mean( Dˆ ) mean( Dˆ ) (8) ˆ, if Sign h (9) W, ele where Ŵ i he -h elemen of waermar equence Ŵ, which i raced from ego-image blindly. From Equaion (9), we now ha he ani-inerference abiliy of hi algorihm lie on he inerval value beween b and a. The larger inerval value i, and he beer ani-inerference abiliy i. 3.3 Informaion Hiding Algorihm Sep : Read a file and conver i ino a bi ream W, which i called he original waermar. Sep : In order o enable ha he lengh of original waermar W i ju equal o 3 ime of oal bloc number of carrier image, ome zero can be appended o he end of i. Sep 3: For improving he robune of waermaring, he redundancy of embedded waermar hould be enured. So he original waermar W hould be ended periodically a follow: m W Wl m n L l; n,,, Cr ; l,,, L (9) m where W i he ended waermar, W repreen i m-h elemen, Cr i he ended facor. Sep 4: In order o improve he ecuriy of waermaring, W hould be crambled randomly. Sep 5: In order o eep he relaiviy of wo neighboring image bloc, we can can he original carrier image by Hilber canning o obain a Hilber canning equence. Sep 6: Selec wo neighboring image bloc B and B each ime from he Hilber canning equence, and embed he waermar according o he mehod a menioned in Secion 3 unil all DWT deail ub-band of all bloc have been proceed. Sep 7: Afer applied he invere DWT for all waermar embedded bloc, we can ge a ego-image I. 3.4 Informaion Recovering Algorihm Sep : Scan he ego-image I by Hilber canning wih he ame order a ha in informaion hiding. Sep : Selec wo neighboring image bloc B and B each ime from he Hilber canning equence, and rac he waermar according o he mehod a menioned in Secion 3 unil all DWT deail ub-band of all bloc have been proceed. Sep 3: Afer ha, we can ge a waermar equence Ŵ, which involve he Cr copie of original waermar. Sep 4: If here wa a crambling when waermar wa embedded, here we hould do uncrambling o Ŵ. Sep 5: The final waermar can be obained from Ŵ a follow: Wˆ l,, if Cr n ele Wˆ nll Cr Sep 6: The binary waermar equence Ŵ be convered bac ino a file. 4. Experimenal Reul () hould The pea ignal-o-noie raio (PSNR) i employed o evaluae he qualiy of ego-image, meanwhile he bi error rae (BER) i employed o evaluae he qualiy of recovered ecre informaion. In he perimen, he propoed algorihm wa evaluaed on he gray image Lena (5 5 8). The bloc ize K can be 8, 4 or. In order o enure he algorihm ani-inerference abiliy i good, [a,b ] hould be e maller and [a,b ] hould be e bigger. So we e [a,b ]=[,]. Figure i he relaionhip beween PSNR of ego-image and he effec facor a differen bloc ize. Figure i he relaionhip beween BER of he recovered ecre bi and he effec facor Figure 3 i he relaionhip beween PSNR and he JND normalized range [a,b ] a full capaciy embedding. Figure. Relaionhip beween PSNR and Figure. Relaionhip beween BER and Copyrigh 9 SciRe

4 Informaion Hiding Mehod Baed on Bloc DWT Sub-Band Feaure Encoding 386 Table. The full capaciie in differenc bloc ize and heir PSNR and BER Capaciy and Performance Hiding capaciy (bi) Hiding capaciy (bye) PSNR (db) BER (%) Figure 3. Relaionhip beween PSNR and a ([a,b]=[,], b=a+) A hown in Figure, he PSNR of ego-image are conan when K =, bu when K >, he bigger i, he higher PSNR are. Bu a nown in Secion 3, he bigger will bring down on he embedding ineniy, i lead ha he performance of recovering algorihm will be wore, epecially when K =8, he BER will be higher. A hown in Figure, we now ha BER are alway zero when, and whaever he K i. So we e he = in he following perimen. A hown in Figure 3, he PSNR are almo in invere proporion o [a,b]. The maller [a,b] i, he higher PSNR are, and he beer impercepibiliy of hidden informaion i. So ha an appropriae [a,b] i beer. Here, we e [a,b]=[6,7]. A hown in Table, he algorihm full hiding capaciie in variou bloc ize and heir performance are good enough o hide informaion. Figure 4(a) i he original image of Lena. Figure 4(b) and 4(c) are he ego-image hidden wih 768 bye when K=8 and he difference image beween he original image and ego-image on he condiion of magnifying 3 ime. Figure 4(d) and 4(e) are he reul when K=4. Figure 4(f) and 4(g) are he reul when K=. A hown in Table and Figure 4, conidering he viual percepion in he algorihm, he hidden informaion in Lena gray image i inviible hough he PSNR i a lile low. The algorihm can fully recover he hidden informaion from he ego-image. (a) (b) (c) 8 8(K=8) Bloc ize 4 4(K=4) (K=) Table 3. The zero BER aac defene e of hor informaion hiding Aac iem Cropping in cenral region Performance under zero BER Brighne enhancemen 69% Conra enhancemen 33% In he perimen, we alo did ome aacing e o he algorihm on he condiion of K=4 and 768 bye wih 8 redundan copie. A hown in Table 3, we found ha he algorihm wa robu o cenral region cropping, brighne enhancemen and conra enhancemen. 5. Concluion Thi paper preened a new cheme of informaion hiding in gray image baed on DWT for long informaion hiding or cover communicaion. In our approach, he comparabiliy of correponding DWT deail ub-band of wo neighboring image bloc wa conidered, and in order o improve he ranparency of informaion hiding, he viual model wa alo ued for calculaing he double bloc baed JND o deermine he embedding ineniy a differen locaion of image. The bloc DWT ub-band feaure encoding echnique increaed he embedding capaciy double han ha of [5]. The adjuable bloc ize gave faciliie for variou applicaion. If you reque he algorihm o have a beer ranparency, you hould elec bigger bloc ize. Or if you reque ha i ha a larger embedding capaciy, you hould elec maller bloc ize. In addiion, in order o improve he algorihm robune and abiliy of defene ome general image proceing aac, uch a cropping, brighne enhancemen and conra enhancemen, he redundan encoding wa given for increaing he embedded copie of waermar. The perimen reul demonrae ha he propoed algorihm yield he accepable performance for ranparency and robune, and alo increae he embedding capaciy for informaion hiding. 6. Acnowledgemen (d) (e) (f) (g) Figure 4. The informaion hidden reul of Lena Copyrigh 9 SciRe Thi reearch projec wa uppored by he Technological Innovaion Foundaion of Shanghai Municipal Educaion Commiion under Gran No. 9YZ456 and he Key

5 Informaion Hiding Mehod Baed on Bloc DWT Sub-Band Feaure Encoding 387 Dicipline of Shanghai Municipal Educaion Commiion under Gran No. J58. REFERENCES [] M. Barni, F. Barolini, and A. Piva, Improved wavelebaed waermaring hrough pixel-wie maing, IEEE Tranacion on Image Proceing, Vol., No. 5, pp ,. [] H. F. Yang and X. W. Chen, A robu image-adapive public waermaring echnique in wavele domain, Journal of Sofware, Vol. 4, No. 9, pp , 3. [3] J. G. Cao, J. E. Fowler, and N. H. Younan, An imageadapive waermar baed on a redundan wavele ranform, In: Pia I, ed. Proceeding of he IEEE Inernaional Conference on Image Proceing, Thealonii, pp. 77 8,. [4] Z. M. Wang, Y. J. Zhang, and J. H. Wu, A wavele domain waermaring echnique baed on human viual yem, Journal of Nanchang Univeriy (Naural Science), Vol. 9, No. 4, pp. 4 43, 5. [5] Q. D. Sun, W. X. Ma, W. Y. Yan, and H. Dai, T encrypion echnique baed on robu image waermaring, Journal of Image and Graphic, Vol. 3, No., pp , 8. [6] L. Y. Wu and F. Yang, An improved digial waermaring algorihm baed on DWT, Conrol and Auomaion, Vol. 3, No. 6(3), pp , 59, 7. Copyrigh 9 SciRe