An Improved Reversible Data Hiding Scheme by Changing Modification Direction of Partial Coefficients in JPEG Images
AAn Improved Reversible Data Hiding Scheme by Changing ModificationDirection of Partial Coe ffi cients in JPEG Images Yi Chen a , Hongxia Wang a, ∗ a School of Information Science and Technology, Southwest Jiaotong University, Chengdu 611756, China
Abstract
This paper first reviews the reversible data hiding scheme, of Liu et al. in 2018, for JPEG images. After that, animproved reversible data hiding scheme, in which modification directions of partial nonzero quantized alternatingcurrent (AC) coe ffi cients are utilized to decrease distortion and file size increase caused by data hiding, is proposed.Experimental results have shown that the proposed scheme has indeed advantages in visual quality and smaller in-crease in file size of marked JPEG images while compared to the state-of-the-art scheme with the same embeddingpayload so far. Keywords:
Visual quality, file size, reversible data hiding (RDH), JPEG images.
1. Introduction
Reversible data hiding (RDH), also called as losslessdata hiding, can embed information in the host, such asimages, the marked host can be restored to the origi-nal host (“clean”) after the embedded information is ex-tracted out. Therefore, it plays a significantly impor-tant role in medical and military fields because of itsreversibility.Up to now, many RDH schemes have been proposedand they are mainly based on histogram shifting (HS)[1], di ff erence expansion (DE) [2] and lossless compres-sion [3]. In fact, there are a fewer researchers who payattention to lossless compression-based RDH schemessince larger embedding payloads cannot be obtained andmore significant degradation in visual quality of markedimages may be caused by the mean of lossless com-pression at present, but we can know most researchesare mainly based on HS and DE according to the lit-erature [4]. Since the technologies of HS and DE arerespectively proposed by Ni et al. and Tian, many im-provements of them have been proposed and designedfor images with di ff erent formats. However, most ofthem are designed for uncompressed images and manysignificantly successful achievements in terms of vi-sual quality and embedding capacity of marked uncom-pressed images have been made in the past two decades. ∗ Corresponding author
Email address: [email protected] (Hongxia Wang)
Hence, designing RDH schemes with good visual qual-ity and embedding capacity for marked compressed im-ages, e.g., JPEG images, has attracted increasingly in-terest from more and more researchers.File size, visual quality and embedding capacity thatare three widely used standards of evaluation must beconsidered when designing RDH for JPEG images. Ingeneral, the requirement of visual quality for encryptedJPEG images is unnecessary and thus it may be easierto design a RDH algorithm for them compared with un-encrypted JPEG images [5]. In other words, designingRDH algorithm for unencrypted JPEG images is morechallenge.Recently, Nikolaidis modified zero quantized alter-nating current (AC) coe ffi cient combined with a map-ping rule to propose a RDH [6]. However, increasingthe number of nonzero quantized coe ffi cients means thatfile size may significantly become larger with the in-crease of embedding capacity. Alternatively, Huang etal. proposed an excellent RDH scheme based on HS forJPEG in [7]. In this scheme, only the AC coe ffi cientswith magnitude 1 are exploited to hide secret informa-tion bits and other nonzero AC coe ffi cients are shifted tovacate room for hiding data. In addition, they only takeadvantage of the blocks with more zero coe ffi cients be-cause it may lead to less invalid shifting and thus obtainhigher visual quality of marked JPEG images. For in-stance, Fig. 1(a) denotes a selected block with quan-tized coe ffi cients for data hiding and it becomes Fig. Preprint submitted to Elsevier May 29, 2018 a r X i v : . [ c s . MM ] M a y a) (b) (c) (d) -23 -3 2 1 0 3 0 00 1 -1 -1 -1 0 0 0-1 1 2 -1 0 0 0 01 1 1 0 0 0 0 0 Figure 1: Coe ffi cient changes before and after data hiding. (a) Orig-inal JPEG quantized coe ffi cients. (b) Marked coe ffi cients by [7] andembedded information bits: “0110 1101 100”. (c) Marked coe ffi cientsby [8] and embedded information bits: “0110 1100 1001 000”. (d)Marked coe ffi cients using the proposed scheme and embedded infor-mation bits: “0110 1100 1001 000”. ffi cient topropose a simple RDH scheme with higher embeddingcapacity for JPEG images. Besides, the scheme ob-tained smaller increase in file size of marked JPEG im-ages compared with that of Huang et al.’s scheme whenusing special encoding principle, which is addressed indetail in [8]. With their scheme, all AC quantized co-e ffi cients are changed from Fig. 1(a) to Fig. 1(c). Toour best knowledge, Liu et al.’s scheme is the state-of-the-art in the embedding capacity at present. However,we are inspired by Liu et al.’s scheme to propose a bet-ter RDH scheme with the same embedding capacity butbetter visual quality and smaller increase in file size ofmarked JPEG images in this paper.
2. The proposed method
The proposed scheme is very simple and the di ff er-ence between it and [8] is that we make full use of an-other modification direction, which is di ff erent from thatof [8], to decrease the increase of file size and keep bet-ter visual quality of marked JPEG image. More detailsare demonstrated in the following. (a) (b) (c) (d) (e) (f) Figure 2: Test images. (a) Lake. (b) Lena. (c) Mandrill. (d) Jetplane.(e) Boat. (f) Elaine.
JPEG images is firstly entropy decoded, divided into8 × ffi cients by the sender. After that, all nonzero quan-tized AC coe ffi cients are utilized to embed data. Thealgorithm of the proposed scheme is stated as follows.¯ C = × C if S = × C − sign ( C ) if S = C and S denote a nonzero quantized AC coe ffi -cient and a to-be-embedded bit, respectively, and sign ( x ) = x > − x < ffi cients willbe changed corresponding to Fig. 1(d). The recipient decodes the marked JPEG images afterhe / she receives them and obtains nonzero quantized ACcoe ffi cients. The embedded information bit is extractedfrom a nonzero coe ffi cient ¯ C and the coe ffi cient of JPEGimages are restored in the following. S (cid:48) = C is even.1 if ¯ C is odd. (3) C (cid:48) = ¯ C / C is even.( ¯ C + sign ( C )) / C is odd. (4)where S (cid:48) and C (cid:48) denote an extracted message bit and arestored AC coe ffi cient, respectively.2 ayload (bits) PS N R ( d B ) Huang et al.'s schemeLiu et al.'s schemeProposed ×10 (a) Lake ×10 Payload (bits) PS N R ( d B ) Huang et al.'s schemeLiu et al.'s schemeProposed(b) Lane
Payload (bits) PS N R ( d B ) Huang et al.'s schemeLiu et al.'s schemeProposed ×10 (c) Mandrill ×10 (e) Boat Payload (bits) PS N R ( d B ) Huang et al.'s scheme
Liu et al.'s scheme
Proposed ×10 (d) Jetplane Payload (bits)0.15 0.30 0.45 0.60 0.75 0.90 1.05 1.20 1.35 PS N R ( d B ) Huang et al.'s schemeLiu et al.'s schemeProposed
Payload (bits) PS N R ( d B ) Huang et al.'s scheme
Liu et al.'s scheme
Proposed ×10 (f) Elaine Figure 3: Comparison of PSNR variations of marked JPEG images (QF = ayload (bits)0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 F il e s i ze ( B y t e s ) Huang et al.'s schemeLiu et al.'s schemeProposed ×10 (a) Lake Payload (bits)0.15 0.30 0.45 0.60 0.75 0.90 1.05 1.20 1.35 F il e S i ze ( B y t e s ) Huang et al.'s scheme
Liu et al.'s scheme
Proposed ×10 (b) Lena ×10 (c) MandrillPayload (bits)0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 F il e S i ze ( B y t e s ) Huang et al.'s schemeLiu et al.'s scheme
Proposed ×10 (d) JetplanePayload (bits) F il e S i ze ( B y t e s ) Huang et al.'s scheme
Liu et al.'s schemeProposed
Payload (bits) F il e S i ze ( B y t e s ) Huang et al.'s scheme
Liu et al.'s schemeProposed ×10 (e) Boat F il e S i ze ( B y t e s ) ×10 (f) ElainePayload (bits)0.15 0.30 0.45 0.60 0.75 0.90 1.05 1.20 1.35 Huang et al.'s schemeLiu et al.'s schemeProposed
Figure 4: Comparison of file size of marked JPEG images (QF = able 1: Comparisons of embedding capacity (bits) Images QF =
50 QF =
70 QF =
3. Experimental results and analysis
With the function of “imwrite” of “MATLAB”, sixstandard 512 ×
512 grayscale images, including “Lake”,“Lena”, “Mandrill”, “Jetplane”, “Boat” and “Elaine”,are converted to JPEG images with di ff erent quality fac-tors, i.e., QF =
50, 70 and 90, shown in Fig. 2, whichare used in experiments to evaluate the performance ofour proposed scheme.We make use of Table 1 to demonstrate the embed-ding capacities of Huang et al.’s scheme, Liu et al.’sscheme and our proposed scheme on these test images.Clearly, Liu et al.’s scheme and our proposed schemecan obtain higher embedding capacity when comparedwith Huang et al.’s scheme. Liu et al.’s scheme andour proposed scheme make full use of each nonzeroAC coe ffi cient to carry secret bits. However, in Huanget al.’s scheme only the nonzero AC coe ffi cients withmagnitude 1 are exploited to carry secret bits and othernonzero AC coe ffi cients are shifted to vacate room forreversibility. Therefore, the first two schemes have in-deed an advantage in the embedding capacity whencompared with Huang et al.’s scheme and their embed-ding capacities are approximately twice as much as thatof Huang et al.’s scheme.Figs. 3-4 are exploited to discuss the visual qual-ity and the increased file size of marked JPEG images.We just give the experimental results on these test im-ages with QF =
50 in this paper and we think they canrepresent other cases, i.e., test images with other dif-ferent quality factors. Obviously, Liu et al.’s schemeand our proposed scheme degradate more significantlycompared with Huang et al.’s scheme according to Fig.3. This is because nonzero AC coe ffi cients are at mostincreased or decreased by 1 in Huang et al.’s schemebut not in our proposed scheme and Liu et al.’s scheme.Furthermore, the visual quality of our proposed schemeis improved when compared with that of Liu et al.’sscheme. Alternatively, we give Fig.4 to compare theincreased file size of marked JPEG images using thethree schemes under the same embedding capacity. Ob- viously, using Liu et al.’s scheme leads to significantincreased file size of marked JPEG images when com-pared with our proposed scheme and Huang et al.’sscheme. From Fig. 4(a-f), we can observe that using ourproposed scheme results in the increased file size closeto that using Huang et al.’s scheme. When there exista lot of nonzero AC coe ffi cients with magnitude of ≤
2, our proposed scheme may have an advantage. For in-stance, if a nonzero AC coe ffi cient is with the value of 2.By Huang et al.’s scheme, it must become 3. Moreover,it will become 4 or 5 which corresponds to the to-be-embedded bit is “0” or “1” with Liu et al.’s scheme. Incontrast, it will become 4 or 3 corresponding to the to-be-embedded bit “0” or “1” by our proposed scheme.That is to say, our proposed scheme can embed onesecret bit sometimes the modification is identical butHuang et al.’s scheme cannot. In result, our proposedscheme leads to the less number of modified coe ffi cientscompared with Huang et al.’s scheme when the embed-ding capacity is same, which makes the increased filesizes very close. Based on the above-mentioned analy-sis, we can know our proposed scheme outperforms Liuet al.’s scheme.
4. Conclusions
We propose a novel and simple reversible data hidingscheme for JPEG images in this paper. Compared withthe state-of-the-art scheme, our proposed scheme can(1) obtain the same embedding capacity. In other words,each nonzero quantized AC coe ffi cient can carry one in-formation bit; (2) improve the visual quality of markedJPEG images; (3) and decrease the increased file size ofmarked JPEG images. In addition, our proposed schemecan keep the increased file size very close to that of thestate-of-the-art HS-based scheme while embedding thesame embedding payloads. In fact, the increased filesize of JPEG images using our proposed scheme is lessthan that of Huang et al.’s scheme when the embeddingpayloads is smaller, e.g., < × bits like Fig. 4(f).5 cknowledgment This work was supported by the National Natural Sci-ence Foundation of China (NSFC) under the grant No.U1536110.
References [1] Z. Ni, Y.-Q. Shi, N. Ansari, W. Su, Reversible data hiding,IEEE Transactions on Circuits and Systems for Video Technol-ogy 16 (3) (2006) 354–362.[2] J. Tian, Reversible data embedding using a di ff erence expansion,IEEE Transactions on Circuits and Systems for Video Technol-ogy 13 (8) (2003) 890–896.[3] J. Fridrich, M. Goljan, R. Du, Lossless data embedding for all im-age formats, in: Security and Watermarking of Multimedia Con-tents IV, Vol. 4675, International Society for Optics and Photon-ics, 2002, pp. 572–584.[4] Y.-Q. Shi, X. Li, X. Zhang, H.-T. Wu, B. Ma, Reversible datahiding: advances in the past two decades, IEEE Access 4 (2016)3210–3237.[5] Z. Qian, H. Xu, X. Luo, X. Zhang, New framework of reversibledata hiding in encrypted JPEG bitstreams, IEEE Transactions onCircuits and Systems for Video Technology PP (99) (2018) 1–1. doi:10.1109/TCSVT.2018.2797897 .[6] A. Nikolaidis, Reversible data hiding in JPEG images utilisingzero quantised coe ffi cients, IET Image Processing 9 (7) (2015)560–568.[7] F. Huang, X. Qu, H. J. Kim, J. Huang, Reversible data hidingin JPEG images, IEEE Transactions on Circuits and Systems forVideo Technology 26 (9) (2016) 1610–1621.[8] Y. Liu, C.-C. Chang, Reversible data hiding for JPEG imagesemploying all quantized non-zero AC coe ffi cients, Displays 51(2018) 51–56.cients, Displays 51(2018) 51–56.