2015_A high capacity data hidingalgorithm for H_264_AVC video

RESEARCH ARTICLEA high capacity data hiding algorithm for H.264/AVC videoMehdi Fallahpour1*,Shervin Shirmohammadi1and Mohammad Ghanbari21School of Electrical Engineering and Computer Science(EECS),University of Ottawa,Ottawa,Ontario,Canada2School of Electrical and Computer Engineering,University of Tehran,Tehran,IranABSTRACTThis article presents an information hiding algorithm for H.264/AVC video stream.It utilizes position of the last nonzero level of quantized discrete cosine transform block to embed information bits.Because only the high-frequency levels are changed,it can guarantee a high peak signal-to-noise ratio and slight increase in bit rate after the watermark embedding. The extraction process is not complex;thus,the proposed technique is an excellent solution for real-time applications such as broadcasting.Experimental results on several test sequences demonstrate that the proposed approach can realize blind extraction with real-time performance;it also provides very high capacity,low distortion and increase in bit rate by about 0.5%.Copyright©2015John Wiley&Sons,Ltd.KEYWORDSvideo data hiding;H.264/AVC;DCT;watermarking;multimedia security*CorrespondenceMehdi Fallahpour,School of Electrical Engineering and Computer Science(EECS),University of Ottawa,Ottawa,Ontario,Canada.E-mail:mfallah@site.uottawa.ca1.INTRODUCTIONWith the progress of computer and broadband internet tech-nologies,nowadays,it is possible to stream high-quality video and to watch video online.Websites such as YouTube,DailyMotion and iFilm provide free watching and download services,whereas portable devices such as iPod,smartphone and tablet make it possible to playback a video anywhere anytime.Thus,digital video has become a popular multimedia content for both online and offline en-vironments.Therefore,it is important to discover ways to protect the contents from malicious use and to design effi-cient methods to search the desired contents in the database, secure ways to provide extra features for upgraded viewers and reliable techniques to recover from transmission error for uninterrupted viewing for the future of digital video.Data hiding methods embed a secret signal directly into the original media in an imperceptible manner.In video data hiding,the watermark can be embedded either into an uncompressed video sequence(raw data)or compressed video stream although the great majority of streamed video through the Internet are compressed,so data hiding in com-pressed video has a wider impact than data hiding in un-compressed video.Transparency,capacity and security are three major challenging issues of a video data hiding system.Robustness is a property of a watermarking system,which is considered in some applications such as copyright pro-tection.In data hiding applications,robustness is not a pri-ority because such applications are more aimed to achieve high transparency along with embedding a great deal of se-cret bits[1].The application of our proposed algorithm is mainly in covert communication and error concealment, but it can also be used in scenarios such as[2]where data hiding techniques are used to measure the quality of the compressed video in the absence of the original reference (such as reduced reference model),by predicting the qual-ity from the degradations of the extracted hidden message. Another application of data hiding can be seen in[3], which uses data hiding to enable real-time scene change detection in compressed video.Finally,data hiding is a suitable solution for error detection and concealment in video transmission,where edge orientation information and number of bits of a block are embedded in the bit stream for that purpose[4].In general,the existing data hiding systems rely on em-bedding information bits in discrete cosine transform (DCT)coefficients,motion vectors(MVs),quantization scale or prediction modes.Li et al.[5]used DCT coeffi-cients and the parity of the quantized coefficients to hide a message.The work in[6]presents a solution for using the magnitude of MVs for data hiding.The least significant SECURITY AND COMMUNICATION NETWORKSSecurity works2015;8:2947–2955Published online17March2015in Wiley Online Library().DOI:10.1002/sec.1221 Copyright©2015John Wiley&Sons,Ltd.2947bit of both components of candidate MVs are used to em-bed a secret message.The candidate MVs are chosen on the basis of the prediction error of the underlying macro-block.MVs related to high prediction errors are selected.A prediction error threshold per frame is calculated and transmitted in the video bit stream to guide the decoder in recognizing the MVs that carry the secret information bits.Hu et al.[7]proposed the use of intraprediction modes to embed secret message bits.Furthermore,the work in[8]utilizes the block types and modes of intracoded blocks of H.264/AVC to hide message bits. Data hiding technique can also be applied before compres-sion,such as embedding hidden messages into the pixels and then compressing them as I-frames[9].Recently,several researchers have focused on devel-oping data hiding systems in the compressed domain [10–16].However,the complex encoding nature of mod-ern encoders such as H.264/AVC[17]creates several constraints for hiding data in the coded video sequences. The restrictions include the following:(1)Embedding even a little data into the coded videosequence would result in a great deal of changeson both the bit rate and the reproduced videoquality.(2)It is difficult to make a cost effective trade-off be-tween the bit-rate increase and video quality degra-dation when embedding secret data into the codedbit stream.(3)To increase hiding capacity,increase in bit rate orvideo quality degradation is sacrificed[18].Because of these constraints,designing a video data hid-ing system in the compressed domain is more difficult than other domains.Two methods[19,20]have applied the tra-ditional reversible methods to H.264/AVC directly.Lie et al.[19]utilize difference expansion[21]to embed secret bits into the quantized discrete cosine transform(QDCT) coefficients for error concealment,whereas Chung et al.[20]suggest an efficient algorithm that utilizes histogram shifting[22]through QDCT coefficients to embed MVs for error concealment,but considering the histogram shifting rules,the system shouldfind the peak and zero points to shift and embed data from the embedding target. According to the context-adaptive variable-length coding (VLC)rule[23],changing the number of QDCT coeffi-cients or the length of the VLC words can significantly increase the bit rate or reduce the peak signal-to-noise ratio(PSNR).In our proposed method,after performing DCT and quantization phases in H.264/AVC,some4×4blocks are selected for embedding.The method takes advantage of high-frequency levels to gain from lower sensitivity of the human visual system to higher frequency distortions.Em-bedding in the low-frequency levels of4×4blocks,which carry perceptually important information,results in obvious quality distortion in the watermarked video.Therefore,in the proposed scheme,the last nonzero(LNZ)QDCT coefficients of4×4blocks are used,which are in the high-frequency band to embed the watermark.The scheme utilizes the positions of the LNZ levels to embed informa-tion bits.Experimental results and analysis presented in Section3will demonstrate that the suggested system pro-vides very high capacity and low-quality degradation while limiting the bit-rate increase to about0.5%.It should be noted that,although in this paper,H.264/ AVC is chosen to demonstrate the proposed method,there is no loss of generality and the method can be applied to any modern video codec,such as the high efficiency video coding(HEVC),which transforms macroblocks using DCT or operations similar to DTC,leading to frequency coefficients.The rest of the paper is organized as follows. In Section2,the proposed data hiding scheme is presented. Section3introduces the experimental results.Finally, Section4gives some concluding remarks.2.THE PROPOSED DATA HIDING SCHEMEAccording to the applications,several video compression standards,such as MPEG-2,Motion Picture Expert Group 4Part2(MPEG-4),Windows Media Video9Codec (VC-1),Motion Picture Expert Group4Part10(H.264 AVC)and On2True-Motion VP6(VP6)can be employed. Almost all video compressions are block DCT based.In these codecs,every frame is divided into blocks where cod-ing operations are applied on DCT coefficients.However, in H.264/AVC like other DCT block-based compression codecs,motion estimation,prediction,entropy coding and compression tricks are applied on each block.The use of DCT coefficients for embedding the water-mark is very challenging in both compressed image and video(such as JPEG and H.264)because compression does not leave much room for watermark location.For the H.264/AVC video codec,the QDCT coefficients and the MVs are the most used parameters for embedding the information bits.Low-frequency,middle-frequency or high-frequency QDCT coefficients may be selected to embed the watermark on the basis of the requirements of the application in terms of robustness or fragility.For the embedding process,the choice of the DCT coefficients usually relies on the human visual properties to mask the introduced degradation by the watermark.In our proposed method,robustness is not a priority;thus,the high-frequency coefficients are chosen because they introduce less perceptual artefacts.The grey area in Figure1shows the locations of high-frequency coefficients suitable for embedding the signature.In H.264/AVC,applying motion estimation,prediction, DCT and quantization results in blocks with value close to one.Because DCT values are quantized,a significant num-ber of QDCT values are converted to zero.In most blocks, there is an LNZ QDCT level.Zigzag scanning for each 4×4block is used for locating LNZ levels.The position of LNZ plays a key role in embedding secret bits in theWatermarking M.Fallahpour,S.Shirmohammadi and M.Ghanbari 2948Security works2015;8:2947–2955©2015John Wiley&Sons,Ltd.DOI:10.1002/secsuggested system.Figure 2provides statistics of position of LNZ for different quarter common intermediate format (QCIF)video tests.This figure illustrates in how many blocks the position of LNZ is equal to P i ,1≤P i ≤16.All positions can be used for embedding secret bits but changing low-frequency and middle-frequency compo-nents may cause signi ficant distortion.Thus,high-frequency levels are used for embedding the watermark.Blue bins in Figure 2show the number of LNZ levels in high-frequency area.2.1.EmbeddingEvery block of 4×4,which at least has a high-frequency coef ficient,can be selected for data hiding.Thr is de fined as a threshold for modi fication of the LNZ position.More speci fically,if the position of the LNZ is greater than Thr ,the block is chosen for embedding,and a single bit isembedded in it.The embedding process for each block af-ter the quantization phase is performed by the following steps:(1)To improve the security,encrypt the embedded wa-termark stream Ws by a key called K ,to form the watermark stream S .S ¼E K ;Ws ðÞwhere E is the encryption operation.(2)For each 4×4block,find the position of the LNZlevel.(3)If the position is higher than Thr ,the block has ahigh-frequency element and is a candidate for embedding,and the embedding process continues.Otherwise return to step 2for the next block.(4)For each selected block,modify the position ofLNZ level (P i )based on S i (the i th bit of the stream)as follows:P ′i ¼P i if P i is even and S i is 0P i þ1if P i is odd and S i is 0P i þ1if P i is even ;P i <16and S i is 1P i if P i is odd and S i is 1P i À1if P i is 16and S i is 18>>>>>><>>>>>>:where S i is the encrypted watermark bit of the i th block.Figure 1.Selected high-frequency area in each 4×4block.50001000015000200002500030000350004000045000ContainerForemanMobileNewsTennis12345678910111213141516Figure 2.Positions of the last nonzero(LNZ)level for five different QCIF test video sequences.Figure 3.Embedding and detecting flowchart.WatermarkingM.Fallahpour,S.Shirmohammadi and M.Ghanbari 2949Security works 2015;8:2947–2955©2015John Wiley &Sons,Ltd.DOI:10.1002/secAfter the aforementioned embedding steps,the block is ready for other steps of the H.264/AVC coding process such as context-adaptive VLC.Figure3shows the embedding and detecting procedures. After decoding the H.264/AVC stream,which includes wa-termark detection,the Luma(Y)and two chrominance (UV)components(YUV)file and the extracted secret bits are obtained.In H.264/AVC,changing the quantization parameter (QP)results in varying the number of non-zero levels in the macroblocks.It is evident that when QP is low,there are more non-zero levels compared with high QP,increas-ing the embedding capacity.2.2.DetectingThe embedded watermark bits are extracted in the decoding process of H.264/AVC where the quantized DCT levels for each MB are entropy decoded.For each block,the following steps result in extracting the embed-ded bits from each block(1)Find the position of the LNZ level.(2)If the position of LNZ is higher than Thr,follow thefollowing steps;otherwise start from step one forthe next block.(3)In the selected block,the embedded bit can beextracted as follows:S i¼0if P′is even 1if P′is odd (where S i is the extracted bit of the i th selected block in the decoder and P′is the position of the LNZ in the current block.To obtain the raw watermark stream,we need to decrypt the extracted stream by the encryption key.In general,efficiency,complexity,energy usage and sim-plicity are more important in the decoder implementa-tion because it is at the client side with limited resources compared with the server side,which has more resources.Also,some delay in the encoder is ac-ceptable because it is carried out only once for a video. So,low complexity and simplicity in implementation are critical points in designing the decoder.One advan-tage of our proposed technique is its simplicity in implementation at the decoder,allowing it to run for various real-time applications.3.EXPERIMENTAL RESULTS AND ANALYSISTo verify the performance of the proposed information hiding system,it has been tested on several test video sequences such as Container,Foreman,Mobile,News and Tennis in QCIF format(176×144pixels),as shown in Figure4.As the watermark is embedded in blocks of 4×4pixels,the embedding algorithm is independent of the resolution of the video.The original video streams are obtained by the H.264/ AVC reference software version JM12.2[24],and the pro-posed method is integrated within the H.264/AVC codec. The most important configuration parameters of the refer-ence software are given in Table I.The rest of the encoder parameters have retained their default values.Also,it is worth to note that in parts A and B,Thr isfixed at9,Figure4.Video test sequences.Watermarking M.Fallahpour,S.Shirmohammadi and M.Ghanbari2950Security works2015;8:2947–2955©2015John Wiley&Sons,Ltd.DOI:10.1002/secwhereas in part C,changing the Thr itself will be discussed.Capacity is defined as the number of bits embedded in one frame or1s of the video.In this article,the capacity is measured and compared with other schemes in terms of embedded bits in a QCIF frame.In an ideal case,there must be no perceptible difference between the watermarked and the original digital contents;that is,the watermark data should be transparent to the user.Embedding in the compressed domain such as H.264/ AVC changes the bit rate of the compressed stream.In most cases,watermark embedding increases the bit rate. The bit-rate increase should be in an acceptable range. For example,the increase in bit rate should not be larger than the watermark bits sent out of band.The trade-off between capacity,transparency and bit rate is the main challenge for video data hiding in the com-pressed domain.In an ideal case,a very high capacity and very transparent scheme without any changes of bit rate are demanded.Practically,achieving all these requirements at the same time is extremely difficult if not impossible. Therefore,depending on the priorities of the data hiding system and the particular application,a trade-off between these properties must be attained.The rest of this section has three parts.In thefirst and second parts,the capacity,bit-rate increase and transpar-ency of the system are presented.The third part provides comparison with the existing methods.3.1.Capacity and bit-rate increaseThe embedding method uses high-frequency LNZ level to hide information.Changing QP varies the number of LNZ levels in the high-frequency area of blocks,where increas-ing QP decreases the number of high-frequency LNZ and results in more compression.Figure5shows how increas-ing QP reduces the amount of embedded bits.It is obvious that when the video sequence is more compressed,there is less vacancy for embedding data bits.Almost any modification in QDCT levels causes the bit rate to increase.Table II presents the original bit rate (H.264/AVC compression without data embedding), marked bit rate(H.264/AVC compression with data embedding)and the effect of the proposed embedding algorithm on the compression process in terms of the dif-ference between the original and marked bit rates.The av-erage capacity is354.8bits per QCIF frame,and the average bit-rate increase is0.47%.3.2.TransparencyPeak signal-to-noise ratio can be used as a measure of transparency of the embedding process.Although PSNR is a very popular and widely used evaluation technique,it is known that its correlation to subjective quality measures is not always good because it reflects only the luminance/ chrominance degradations and neglects the behaviour of human perception.Perhaps,a more reliable measuring de-vice is the Structural Similarity Index(SSIM)[25]. Whereas PSNR measures errors between the original and marked image,SSIM measures the structural distortion,lu-minance and contrast differences between the two frames. The main idea behind SSIM is that the human vision sys-tem is specialized in extracting structural information from the viewingfield,and it is not specialized in extracting the errors.Therefore,a measurement on structural distortion can give a better correlation to subjective impressions. Many different quality assessment techniques canbeFigure5.Capacity and QP forfive video test sequences.Table I.Configuration parameters of the JM software.Profile BaselineLevel3Intra period1Frame rate30Rate distortion optimization OnNumber of encoded frames100Watermarking M.Fallahpour,S.Shirmohammadi and M.Ghanbari2951 Security works2015;8:2947–2955©2015John Wiley&Sons,Ltd.DOI:10.1002/secT a b l e I I .C a p a c i t y a n d b i t r a t e (k b p s )f o r a v e r a g e o f 100f r a m e s f o r fiv e v i d e o s e q u e n c e s w i t h d i f f e r e n t q u a n t i z a t i o n p a r a m e t e r (Q P ).Q P =24Q P =34Q P =44O r i g i n a lM a r k e d D i f f e r e n c e (%)O r i g i n a l M a r k e d D i f f e r e n c e (%)O r i g i n a l M a r k e d D i f f e r e n c e (%)C o n t a i n e rB i t r a t e 1146.6891153.3010.58462.9288467.35680.96157.956158.280.21C a p a c i t y 591.34226.8110.51F o r e m a nB i t r a t e 1133.151141.0420.70429.876433.31760.80170.1432170.3270.11C a p a c i t y 498.58116.037.01M o b i l eB i t r a t e 3165.2813174.2570.281570.4161580.1820.62509.6136510.8040.23C a p a c i t y 1328.22890.1790.52N e w sB i t r a t e 1124.591131.9480.65508.5576511.40640.56197.5296197.70.09C a p a c i t y 405.62131.168.77T e n n i sB i t r a t e 1376.2011382.6810.47449.4024452.16720.62157.0344157.38480.22C a p a c i t y 807.72182.8527.58A v e r a g eB i t r a t e 1589.1821596.6460.536684.2362688.8860.712238.4554238.8610.17C a p a c i t y 726.296309.40428.876T a b l e I I I .T r a n s p a r e n c y i n t e r m o f p e a k s i g n a l -t o -n o i s e r a t i o (P S N R ;i n d B )a n d S t r u c t u r a l S i m i l a r i t y I n d e x (S S I M )f o r a v e r a g e o f 100f r a m e s f o r fiv e v i d e o s e q u e n c e s w i t h d i f f e r e n t q u a n t i z a t i o n p a r a m e t e r (Q P ).Q P =24Q P =34Q P =44O r i g i n a l M a r k e d D i f f e r e n c e O r i g i n a l M a r k e d D i f f e r e n c e O r i g i n a l M a r k e dD i f f e r e n c eC o n t a i n e rP S N R 39.4836.662.8232.3731.251.1325.7125.640.07S S I M 0.957720.950550.007170.897810.891410.006400.783970.782690.00128F o r e m a nP S N R 39.3537.861.4932.2531.800.4525.7825.680.10S S I M 0.969260.965810.003450.904130.901300.002830.737570.736580.00099M o b i l eP S N R 37.9031.406.5028.7326.202.5221.0920.840.25S S I M 0.989330.978320.011010.928080.904820.023260.671690.662420.00927N e w sP S N R 40.4338.811.6232.7232.110.6125.4725.410.06S S I M 0.981160.978530.002630.929170.926770.002400.766710.766390.00032T e n n i sP S N R 38.5535.592.9631.5930.640.9526.1825.990.19S S I M 0.948270.935890.012370.795080.788370.006710.667470.664550.00292A v e r a g eP S N R 39.14236.0643.07831.53230.41.13224.84624.7120.134S S I M 0.9691480.961820.0073260.8908540.8825340.008320.7254820.7225260.002956WatermarkingM.Fallahpour,S.Shirmohammadi and M.Ghanbari2952Security works 2015;8:2947–2955©2015John Wiley &Sons,Ltd.DOI:10.1002/secestablished from this assumption but Wang et al.suggested a simple and effective index algorithm [25].Structural Similarity Index is in the range of 0–1,where 0shows zero correlation (i.e.,the reference frame is entirely different from the target)and 1indicates that they are iden-tical.Table III presents the PSNR and SSIM for 100frames of each test sequence with three different QP values (QP =24,34,44).The last two rows are the averages of the five sequences for PSNR and SSIM.The overall aver-age results for all three QPs show a quality degradation of 1.4dB in terms of PSNR and 0.0061for SSIM.The extraction process at the decoder is not complex,which makes the scheme a perfect solution for real-time applications.3.3.Changing thresholdTo show the effects of changing the threshold,Thr ,capac-ity and transparency are measured while Thr is set to 7,9and 11.As Table IV shows,when Thr is equal to 7,there are more LNZ positions changing,which results in more capacity and more distortion compared with when Thr is 9.Moreover,when Thr is equal to 11,there are fewer LNZ positions changing,which leads to less capacity and better transparency compared with when Thr is 9.As such,we can say that there is an inverse relationship between Thr and capacity,whereas there is a direct relationship between Thr and transparency.parison with the existing methods As each data hiding method has its own special properties under different conditions,it is dif ficult to provide a very accurate comparison.Nevertheless,the proposed method can be compared with some existing methods as explained as follows.Yang et al .[8]proposes an information hiding algo-rithm based on intra-prediction modes and matrix coding for H.264/AVC video stream.Yang et al .[8]improves the data hiding algorithm proposed by Hu et al.[26].Yang et al .[8]and Su et al .[26]provide low distortion;however,their capacity is very low,and also,the bit-rate increase is considerable.Ma et al .[27]introduces a high-capacity scheme,which increases the bit rate dramati-cally.Lin et al .[9]enhances the scheme of Ma et al .[27]in terms of capacity and bit-rate increase,but it does not provide general results and only compares its scheme with [27].The original capacity,PSNR and SSIM (with-out embedding)are not reported in [9],so it is really dif-ficult to know the ef ficiency of the system.Mansouri et al.[28]present a high-capacity method,whereas the bit-rate increase is considerable and transparency is not reported.As Table V shows,simulated outcome in a wide range of QPs,24–44.The capacity of the proposed method is very high,and although [27]provides more capacity,the bit-rate increase in [27]is about seveb times more than the proposed method.Noorkami and Merereau [29]and Gong and Lu [30]provide good capacity,but the bit-rate increase is five and eight times more than ours,respec-tively.Also,Noorkami and Merereau [29]distort the im-age by about 6.5dB,which may not be acceptable for many applications.Gong and Lu [30]reduces the quality of the frames by about 3.7dB,which is about three times more than the proposed algorithm.Our suggested scheme presents the best bit-rate in-crease,less than half of the best reported method.On the basis of the SSIM results,our proposed method is veryTable IV.Capacity and transparency for average of 100frames for five video sequences with different thresholds whereQP =24.Thr 7911Container Capacity 754.96591.34334.77PSNR 35.4436.6638.81Foreman Capacity 725.33498.58224.22PSNR 36.7637.8639.2Mobile Capacity 1492.661328.221060.56PSNR 28.0831.435.68News Capacity 549.64405.62311.16PSNR 37.638.8139.15Tennis Capacity 993.76807.72709.04PSNR 34.9335.5936.45AverageCapacity 903.27726.296527.95PSNR34.56236.06437.858PSNR,peak signal-to-noise ratio.Table parison in terms of capacity,bit-rate increase and transparency.Method QP Capacity bits per frameBit-rate increase (%)PSNR degradation SSIM degradation [8]Not reported33–42 1.60.02Not reported [26]2825–40 2.80.02Not reported [27]28840 3.6 2.1Not reported[9]18–43Not reported1.68~6Just marked SSIM is reported [28]Not reported 274 1.1Not reportedNot reported [29]Not reported 764 3.8 6.5Not reported [30]Not reported 766 2.3 3.7Not reportedProposed24–443550.471.40.0061QP,quantization parameter;PSNR,peak signal-to-noise ratio;SSIM,Structural Similarity Index.WatermarkingM.Fallahpour,S.Shirmohammadi and M.Ghanbari 2953Security works 2015;8:2947–2955©2015John Wiley &Sons,Ltd.DOI:10.1002/sectransparent.However,because other schemes do not pro-vide SSIM,comparison in terms of PSNR is also provided in the table.4.CONCLUSIONSTo design an efficient and low complexity method,the em-bedding and extracting of watermarks are integrated with the coding and decoding routines of H.264/AVC.The in-formation hiding system utilizes position of the LNZ level of QDCT block to embed the information bits.To assure transparency to the human visual system,the high-frequency levels are changed.Changing just high-frequency levels can guarantee a high PSNR,SSIM and slight bit-rate increase after the watermark embedding.As the extraction procedure is not complex,the pro-posed technique is an excellent solution for real-time appli-cations such as broadcasting.Experimental results on several test sequences prove that the proposed approach can realize blind extraction with real-time performance,de-livers very high capacity and low distortion and also keeps the bit-rate increase by about0.005. REFERENCES1.Stefan Winkler EDG,Ebrahimi T.Toward perceptualmetrics for video watermark evaluation.SPIE2003;5203:371–378.2.Carli M,Farais M,Gelasca ED,Tedesco R,Neri A.Quality assessment using data hiding on perceptually important areas.In Proc.IEEE Int.Conf.Image Processing,ICIP.Magazzini de Cotone Modulo9 Genova:Italy,2005;III-1200-3–III-1200-3.3.Kapotas S,Skodras A.A new data hiding scheme forscene change detection in H.264encoded video se-quences.In Proc.IEEE Int.Conf.Multimedia Expo ICME.Hanover,Germany,2008;277–280.4.Yilmaz A,Aydin A.Error detection and concealmentfor video transmission using information hiding.Sig-nal Processing:Image Communication2008;23(4): 298–312.5.Li Y,Chen H-X,Zhao Y.A new method of data hid-ing based on H.264encoded video sequences.In Proc.IEEE Int.Conf.Signal Processing,ICSP.Beijing, China,2010;1833–1836.6.Aly HA.Data hiding in motion vectors of compressedvideo based on their associated prediction error.IEEE Transactions on Information Forensics and Security 2011;6(1):14–18.7.Hu Y,Zhang C,Su rmation hiding based onintra prediction modes H.264/AVC.In Proc.IEEE Int.Conf.Multimedia and Expo,ICME.Beijing, China,2007;1231–1234.8.Yang G,Li J,He Y,Kang Z.An information hiding al-gorithm based on intra-prediction modes and matrix coding for H.264/AVC video stream.AEU Interna-tional Journal of Electronics and Communications 2011;65(4):331–337.9.Lin T-J,Chung K-L,Chang P-C,Huang Y-H,Liao H-YM,Fang C-Y.An improved DCT-based perturbation scheme for high capacity data hiding in H.264/AVC intra frames.Journal of Systems and Software2013;86:604–614.10.Lin YC,Li JH.Data hiding with rate-distortion optimi-zation on H.264AVC Video.In Pro.23th IPPR Conf.on Computer Vision,Graphics,and Image Processing, CVGIP.Kaohsiung,Taiwan,2010;15–17.11.Huang X,Abe Y,Echizen I.Capacity adaptive syn-chronized acoustic steganography scheme.Journal of Information Hiding&Multimedia Signal Processing 2010;1(2):72–90.12.Lin SD,Chuang CY,Chen MJ.A WatermarkingScheme by Changing the Number of Nonzero Coeffi-cients in H.264/AVC Encoder.In Pro.22th IPPR Con-ference on Computer Vision,Graphics,and Image Processing,CVGIP,Aug.23–25,2009.13.Jiang X,Sun T,Zhou Y,Shi Y.A drift compensationalgorithm for H.264/AVC video robust watermarking scheme.Digital Forensics and Watermarking2012;7128:30–41.14.Tang Jian,Zheng Jianfeng,Guo Li.H.264/AVC com-pressed domain data hiding algorithm based on in-loop compensation.Image and Signal Processing(CISP), 20114th International Congress on.Vol.1.IEEE,2011.15.Mehmood N,Mushtaq M.Blind watermarking schemefor H.264/AVC based on intra4×4prediction modes.Future Information Technology,Application,and Ser-vice2012;179:1–7.16.Huo W,Zhu Y,Chen H.A controllable error-driftelimination scheme for watermarking algorithm in H.264/AVC stream.IEEE Transactions on Signal Pro-cessing2011;18(9):535–538.17.Richardson IEG.H.264and MPEG-4Video Compres-sion:Video Coding for Next Generation Multimedia.John Wiley&Sons Ltd:New Jersey,2003.18.Shahid Z,Chaumont M,Puech W.Considering the re-construction loop for data hiding of intra and inter frames of H.264/AVC.In Pro.European Signal Processing Conf.,EUSIPCO.EURASIP:Glasgow,Scotland,2009.19.Lie WN,Lin TCI,Tsai DC,Lin GS.Error resilientcoding based on reversible data embedding technique for H.264/AVC video.In Proc.of IEEE International Conf.on Multimedia and Expo,ICME.Amsterdam, The Netherlands,2005.20.Chung KL,Huang YH,Chang PC,Liao HYM.Re-versible data hiding based approach for intra frameWatermarking M.Fallahpour,S.Shirmohammadi and M.Ghanbari 2954Security works2015;8:2947–2955©2015John Wiley&Sons,Ltd.DOI:10.1002/sec。