A Robust Video Watermarking Algorithm in H.264/AVC Compressed Domain Lotfi ABDi
Faten Ben Abdallah
Aref Meddeb
National Engineering School of Tunis University of Tunis El Manar Tunis, Tunisia
National Engineering School of Sousse University of Sousse Sousse, Tunisia
National Engineering School of Sousse University of Sousse Sousse, Tunisia
[email protected] [email protected] [email protected] ABSTRACT
Keywords
authentication by applying a watermark directly to the entropy coded H.264/AVC stream. The watermark data is embedded and entropy-coded, which modifies the H.264/AVC bit stream. In [2], a robust video watermarking method with a blind extraction process is proposed. The watermark data is embedded into the P-frames of the H.264/AVC video stream, but the resulting bit rate increase is quite high. In [1], a robust video watermarking for H.264 which is based on motion coherence is proposed. The watermark is embedded into the P-frames of the compressed video stream, but the bit rate increase is quite high. In this paper, an efficient algorithm to add a watermark for real time application is presented for the H.264/AVC codec. We refer to our scheme as Real-Time Video Watermarking Algorithm (RTWA). The proposed scheme is robust against re-encoding and offers consistent payload capability to the H.264/AVC standard at different bit rates, without adversely affecting the overall bit rate and the Peak Signalto-Noise Ratio of the video bit stream. The remainder of the paper is organized as follows: section II describes the proposed method by elaborating its embedding and extraction steps. In section III we describe the experimental results and analysis. Finally, section IV draws some conclusions future research directions.
H264/AVC, Video watermarking, Compressed domain, Nonzero quantized transform coefficients.
2.
In this paper, a low complexity video watermarking scheme for H.264 has been presented. Our contribution is to attain lower complexity in embedding procedure and extracting watermark. At the same time, we avoid a bit-rate increase and improve the runtime-efficiency and embedding capacity without sacrificing quality. The watermark is embedded into a video sequence by modifying the number of nonzero-quantized AC coefficients in a 4x4 block of I frames. The experimental results show that the proposed method can prevent a bit-rate increase and improve the runtimeefficiency and embedding capacity without sacrificing the perceptual quality.
Categories and Subject Descriptors H.4 [Multimedia and Visualization ]: Security and Protection, Authentication process.
General Terms Security, Integrity, Algorithms.
1.
PROPOSED METHOD
In this section, we describe the embedding and detecting watermark procedures. By using the advantages of compressed domain, we analyze the syntactic elements of the H.264/AVC coding standard, whose quantized I-frame steps (Qstep) are smaller than B and P frames; i.e., B and P frames are quantized more heavily than I frames [4]. Thus, there is a more redundant space available to embed watermark in I frames. We will illustrate that the watermarking scheme we propose is robust and exhibits a lower complexity of the embedding and extracting processes. This scheme is suitable for real-time encoders used for video conferencing, IPTV and video surveillance systems. The embedding and extraction algorithm in proposed in this section.
INTRODUCTION
Digital watermarking is an excellent tool to ensure security and protection of multimedia data by embedding some information into the digital production. The watermark can be added either into the uncompressed or compressed video. The H.264/AVC video coding standard has become the most widely deployed video codec in almost all applications. As such, various watermarking methods tailored to the H.264/AVC are proposed. For example, Zou and Bloom [5] proposed a blind watermarking algorithm designed for Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. SAC’15 April 13-17, 2015, Salamanca, Spain. Copyright 2015 ACM 978-1-4503-3196-8/15/04...$15.00. http://dx.doi.org/10.1145/2695664.2695989
2.1
Embedding Algorithm
In order to guarantee visual imperceptibility, watermark data should be embedded into more textured blocks. The residual block is a more textured region if there are more nonzero quantized coefficients. Therefore, the number of nonzero (NNZ) quantized coefficients can be considered to estimate the texture of residuals.
1291
For the embedding process, the 4x4 block in which the value of NNZ is no less than threshold K are selected. After quantization, the number of nonzero AC coefficients in this block is counted. Then, the nonzero AC coefficient is marked as NZACi, where i is the order of this NZAC coefficient by ZigZag scan and a secret key decides which coefficients NZACi will be embedded in the block as follows:
Table 1: Comparing between original video and watermarked sequences for QP=28, using PSNR, SSIM, and BIR. Video Embedding SSIM PSNR BIR Size capacity % decrease 10−3 (bits) (dB) % ” Container” 1325 0.999 0.07 0.125 ”Mother” 1268 0.999 0.11 0.165 ”News” 2017 1 0.07 0.0448 ” Hall” 1243 0.999 0.19 0.06 ” Mobile” 1268 1 0.02 0.0116 ”Stefan” 1562 0.999 0.09 0.0135
if Wi = 0 ACi if 0=N ZACi (mod 2) ACi + 1 if 1=N ZACi (mod 2) and NZACi > 0 ACi = AC − 1 if 1=N ZAC (mod 2) and NZACi < 0 i i if Wi = 1 ACi if 1=N ZACi (mod 2) ACi + 1 if 0=N ZACi (mod 2) and NZACi > 0 ACi = AC − 1 if 0=N ZAC (mod 2) and NZACi < 0 i i (1)
experiment allows justifying the invisibility and ro-bustness of our approach against various attacks that modify the inserted information. In the first step, we are interested in the Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Metric SSIM measures to conclude about the quality of the video after embedding the watermark and the Bit-rate Increase (BIR). In the second step, the technique robustness is tested against recompression. Finally, we compare the pro-posed approach with previous works.
Our proposed scheme takes advantage of the H.264/AVC codec to embed the secret information. In addition, the entropy decoding and encoding are two fast procedures allowing signature insertion and detection in real time. Further, the mark insertion in these coefficients improves significantly the robustness of the watermarking algorithm and increases the energy of the mark, without a visible deterioration of the processed signal. The block selection using a pseudo random key enhances the security of the proposed method.
2.2
3.1
Extraction Algorithm
During the watermark extraction procedure, we use exactly the reverse process of watermark embedding to find the watermarked coefficients. The watermark extraction procedure is performed between entropy decoding and inverse quantization. First, the decoder has to be informed about the secret key, which can locate the 4x4 embedded block k according to the secret key (the values of NNZ no less than threshold K are selected). Second, the decoder selects the coefficients NZACi in the I-4x4 blocks. If it is an odd number, the watermark bit is 1 , otherwise it is 0. If a H.264/AVC stream receiver suspects that the video stream is tampered with or intentionally modified for any reason, the watermark extraction and verification algorithm can be applied to confirm the authenticity and integrity. The main steps of the watermark extraction and verification are as follows: Wi =
0 1
if 0 = N ZACi (mod 2) if 1 = N ZACi (mod 2)
BIR =
BR − BR ×100 BR × Embadded Capacity
(3)
Where BR and BR are the number of bits for coding the original and the watermarked sequences, respectively. The value of PSNR, SSIM, embedded capacity, and BIR(x10−3 ) are provided in Table 1. From Fig. 1, we can see that the PSNR of the embedded frames are slightly different from the original frames. Our scheme requires significantly less distortion, about 0.09 dB, while the SIMM values are all above 0.999 and the average BIR is less than 0.08 10−3 %. It is almost impossible to detect the degradation in video quality caused by watermark embedding, which demonstrates that watermark embedding introduces little influence on the video quality. This is mainly because of the small differences between the watermarked video sequences and the reference.
(2)
Obviously, the extraction process is simple and fast because the hidden authentication information can be detected solely from the nonzero AC residuals and can be easily accessible in intra/inter prediction modes. The selection of appropriate blocks, the embedding algorithm, and the use of a pseudo random key yield a low impact on bit-rate and enhance security.
3.
Imperceptibility after watermarking
To estimate visual imperceptibility of the watermark embedding algorithm, the PSNR is usually taken to evaluate the perceptual quality. However, for many video contents, the PSNR cannot be a reliable method for assessing the video quality. Another metric objective is Structural Similarity Index Metric (SSIM). The SSIM index lies in the range of [-1,1], where -1 indicates zero correlation and 1 indicates identical frames [3]. Since the bit-rate increase is related to the embedded capacity, we define the Bit-rate Increase Ratio (BIR) as the percentage of the bit-rate increase per embedded bit.
3.2
Robustness against recompression
Robustness is the resistance of an embedded watermark against intentional attack and normal signal processing operations, whereas the re-compression is the most traditional non intentional attack against watermarked videos. For testing the robustness against re-encoding, we attack our video under re-encoding. We define the Bit Error Rate (BER) as the frequency of bit errors when detecting a multi-
EXPERIMENTAL RESULTS
The proposed method is implemented using H.264/ AVC JM 17.4 software, along with the CAVLC entropy coding with Quantization Parameter (QP) value set to 28. This
1292
Table 3: Coding time of different video sequences. Video Original Watermarked Sequence Sequence (sec) Sequence (sec) ”Container” 279.954 280.522 ”Mother” 305.156 305.363 ”News” 293.863 294.137 ”Hall” 339.552 340.080 ”Mobile” 303.112 303.471 ”Stefan” 355.534 355.667
44 PSNR of original frame(dB) PSNR of watermarked frame(dB)
PSNR (dB)
42
40
38 previously proposed methods, which lead to an important BIR for the watermarked video.
36 20
22 24 26 28 Quantization Parameter(QP)
4.
30
Figure 1: The PSNR before and after watermarking for the ”container” sequence.
bit watermark message [2]. BER =
N umber of error bits T otal number of bits sent
(4)
Table 2 illustrates the cross-correlation peaks for four benchmark video sequences. These results show that the water-
5.
Table 2: Watermark detection rate under reencoding at different QP values (originally encoded at QP value of 28). Video QP 28 30 32 34 ”Container” 0.98 0.81 0.77 0.71 ”Mother” 1.00 0.95 0.81 0.76 ”News” 0.98 0.89 0.74 0.77 ”Hall” 1.00 0.91 0.82 0.78 ”Mobile” 1.00 0.92 0.87 0.79 ”Stefan” 0.99 0.86 0.72 0.69
CONCLUSIONS
In this paper, we proposed a robust Real-Time video Watermarking Algorithm (RTWA) for the H.264/AVC in the compressed domain, with a blind extraction process where original video data is not required to retrieve the embedded watermark. To design an efficient and low complexity method, the embedding and extracting of watermarks are integrated with the coding and decoding routines of the H.264/AVC. The extensive experimental results have demonstrated that the proposed scheme can prevent bit-rate increase and improve embedding capacity without sacrificing perceptual quality. The selection of appropriate blocks, the embedding algorithm, and the use of a pseudo random key yield a low impact on bit-rate and enhance security.
ACKNOWLEDGMENTS
This work was performed under the MOBIDOC device, part of the Support to Research and Innovation System project (PASRI), funded by the European Union (EU) and administered by the National Agency for Promotion of Scientific Research (ANPR).
6.
REFERENCES
[1] T. Dutta. Motion compensated compressed domain watermarking. In Proceedings of the 21st ACM international conference on Multimedia, pages 1039–1042. ACM, 2013. mark is successfully detected and indicates that the algo[2] T. Dutta, A. Sur, and S. Nandi. A robust compressed rithm has a low vulnerability. The lowest correct rate is domain video watermarking in p-frames with controlled still above 69 %. The experimental results given in Table bit rate increase. In Communications (NCC), 2013 2. demonstrate that the proposed scheme is able to achieve National Conference on, pages 1–5. IEEE, 2013. acceptable robustness against [3] Q. Li and R. Wang. Watermarking in h. 264/avc compressed domain using cavlc. Journal of Computers, 3.3 Real-time performance 8(12), 2013. As the watermark is embedded during H.264 encoding, [4] Z. Liu, G. Cheung, and Y. Ji. Unified distributed the embedding process will influence the video coding speed. source coding frames for interactive multiview video The presented technique avoids full decoding and re-encoding streaming. In Communications (ICC), 2012 IEEE in both embedding and extracting phases. It exhibits low International Conference on, pages 2048–2053. IEEE, computational complexity and can meet the requirement of 2012. real time performance such as IP-TV and digital TV broad[5] D. Zou and J. A. Bloom. H. 264/avc stream casting. Results related to the Real-timing are provided in replacement technique for video watermarking. In Table 3. Acoustics, Speech and Signal Processing, 2008. ICASSP The process of embedding and extraction should be lightweight 2008. IEEE International Conference on, pages to respond in an appropriate time, it can meet the require1749–1752. IEEE, 2008. ment of real-timing. We can notice that our scheme is very suitable for real-time and not influence the video encoding speed greatly. This is a significant advantage compared to
1293
