Science, Universiti Brunei Darussalam Abstract: This paper proposes

A Proposed Robust Video Watermarking Algorithm: Enhanced Extraction from Geometric Attacks R. Maharajan1, Abeer Alsadoon1, P.W.C. Prasad1, A. M. S. Rahma2, A. Elchouemi3, SMN Arosha Senanayake4 1School of Computing and Mathematics, Charles Sturt University, Sydney, Australia 2Computer Science Department, University of Technology, Baghdad, Iraq 3Hewlett Packard Enterprise 4Faculty of Science, Universiti Brunei Darussalam Abstract: This paper proposes a new integrated system based on a video watermarking algorithm with high imperceptibility, improved security, robust against common image and video processing attacks and geometric attacks. The proposed algorithm is based on the features of a selected best algorithm by Agilandeeswari and Ganesan and its limitations. The advantages of the hybrid transform techniques of Nonsubsampled Contourlet Transform, Discrete Wavelet Transform and Singular Value Decomposition are utilized for high robustness against common attacks. The features of angle invariance and distance invariability of Log Polar Transform and Inverse Log Polar Transform are used in the extraction process only to resist geometric attacks without affecting the imperceptibility of watermark and watermarked video. Modified Arnold Transformation is introduced to improve the security of the watermark. The experiment shows that the algorithm is extremely robust in terms of most common attacks and geometric attacks with high imperceptibility and improved security. Keywords— Video Watermarking; Authentication; Robustness; Imperceptibility; Payload; Security; Key Frames; Non Key Frames; Modified Arnold Transform; I. INTRODUCTION The fast paced advancement in multimedia and internet technology has significantly enhanced the level and speed of knowledge exchange. Due to this evolution, people nowadays can easily share or watch videos through the internet [1]. However, this evolution has equally generated copyright and authentication issues with videos. Therefore, the need for protecting the authenticity and integrity of videos is the focus of this research. Digital Video Watermarking Technique is one of the solutions for copyright protection which is the process of embedding digital watermarks so as to hide the copyright information in the video and protect its ownership [1]. According to [2], the digital watermarking process has certain factors such as robustness, imperceptibility, security, capacity and computational time that are required to be considered. There are several video watermarking schemes proposed by researchers focusing on providing high robustness against image and video processing attacks, temporal attacks and geometric attacks. It is a significant challenge to provide robustness against almost all kinds of attacks specially in the case of geometric attacks. Hence, most of the existing algorithms have failed to provide robustness against one type of attack while focusing on others. The algorithm proposed by [3], has provided high levels of robustness against most attacks. Moreover, this algorithm has provided high imperceptibility, dual security and high payload. However, this algorithm has some limitations regarding robustness against geometric attacks and security provided by traditional Arnold Transform. To overcome these limitations, the proposed algorithm focuses on providing robustness against geometric attacks without affecting other types of attacks or imperceptibility and improves security by introducing modified Arnold Transformation. This paper is organized as follows: Section I introduces the project, and section II reviews existing research into face detection. This is followed by a description of the methodology used in the proposed solution, in section III, together with an analysis and overview of implementation strategies. Results of the current research into the proposed solution are discussed in section IV and results and discussions are presented in section V. The conclusion is covered under section VI, which also considers possible future work. II. RELATED WORK A significant amount of research has been carried out related to video watermarking. In order to provide an effective and robust video watermarking algorithm, [4] focused on distortion resistance based on an additive spread spectrum and a periodic watermark concept to protect against piracy of the video files. However, it is less resistant to noise attacks and also the use of the spread spectrum is computationally complex. [5] proposed a new video watermarking algorithm based on Singular Value Decomposition (SVD) and slopebased embedding technique focusing on temporal dimension attacks. However, this algorithm has provided less robustness against most of the common attacks. Similarly, [6] fused dual transform domains 2D Discrete Wavelet Transform (DWT) and 3D Discrete Cosine Transform (DCT) with Particle Swarm Optimization (PSO) in order to improve the robustness of the watermark in the video watermarking. [7] introduced an adaptive video watermarking using a human visual system with a fuzzy interference model to provide high robustness and imperceptibility. However, it is less resistant to geometric attacks. The algorithm proposed by [8] used shot boundary detection which reduces the computational time and provides robustness against temporal attacks and classifying blocks of the compressed video. However, this method is less robust against geometric attacks and other noise attacks. [3] proposed a robust video watermarking scheme using hybrid techniques of Contourlet Transform (CT) and DWT for enhancing the robustness and the visual perception. This algorithm has maintained the security using Arnold transform and has achieved high imperceptibility and payload using a bit plane slicing. However, CT basically lacks the property of the shift invariance. Furthermore, the Arnold transformation used in this scheme is the traditional one which is less secure. [9] proposed a robust watermarking scheme based on nonsubsampled contourlet transform (NSCT) which is multidirectional as well as shift invariant. However, the robustness against geometric attacks has not significantly improved. Among all the image and video processing attacks, the most challenging attacks to be considered when providing a robust video watermarking algorithm are geometric attacks, also known as RST (Rotation, Scaling and Translation) attacks. The schemes presented by [10] and [11] have focused on robustness against geometric attacks using Discrete Fourier Transform (DFT) and Log Polar Mapping (LPM) for images. The algorithm used for images can also be applied for video frames. [10] focused on providing a RST invariant watermarking for image using approximate Inverse LPM to get the location to be the watermarked image. However, this scheme also had drawbacks in the form of ILPM which introduces interference distortion. To overcome this drawback, [11] introduced a new scheme which reduces the distortion caused by ILPM. However, this scheme failed by not being able to extract in the case of fractional angles of rotation attacks. To reduce this limitation, [12] presented color watermarking for image using Fourier transform and improved ULPM. However, improved inverse LPM still provides distortion effects. Furthermore, although these schemes focused on geometric attacks, they are all based on DFT which is less robust and more complex than other transform domains. Similarly, to provide robustness against RST attacks, Wang et al. [13] proposed a compressed video watermarking algorithm focusing on resistance to geometric attacks with invariance of a Histogram Shape in DWT Domain and the real time performance by using a fast inter transformation between Block DCTs and One-Level DWT. However, it is less resistant against temporal attacks and the video quality is also degraded. Likewise, to achieve high robustness against geometric attacks, [14] has used a zero watermarking algorithm based on PM with 2D DWT and 3D DCT. However, this algorithm is fixed for authentication only rather than extracting the original watermark. From the above research, it is
evident that achieving high robustness against geometric attacks while maintaining the imperceptibility and robustness against other common attacks represents one of the most challenging features of any video watermarking algorithm. Most schemes failed to consider geometric attacks whereas others did not maintain the imperceptibility in the case of other common attacks. Most of these proposed solutions have focused on robustness of the algorithm and to a much lesser extent on security which, however, is also of high importance. I. PROPOSED WORK A Proposed Robust Video Watermarking Algorithm: Enhanced Extraction from Geometric Attacks (PRVWA-EEfGA) is based on the existing video watermarking algorithm by [3] with reduction of its limitations. The proposed system has utilized the features of this existing algorithm by using scene change detection for reducing computational time, hybrid transform techniques for the purpose of high robustness against image and video processing attacks, temporal attacks and other common attacks, bit plane slicing in watermark for high imperceptibility and capacity and finally Arnold Transform and Eigen Vector for dual security. To reduce the limitation caused by lack of robustness against geometric attacks, the proposed algorithm utilizes a scaling invariance feature of Nonsubsampled Contourlet Transform in the embedding process and distance and angle invariability features of Log Polar and Inverse Log Polar transforms in the extraction process. Due to the interpolation and discretization property of Log Polar Transform and distortion property of Inverse Log Polar Transform, this system significantly enhances robustness against major geometric attacks of rotation, scaling and transition. To improve the limitation of less secure traditional Arnold Transform, modified Arnold Transform is introduced. There are two main stages in this algorithm, the embedding stage and the extraction stage. Fig. 1. Proposed PRVWA-EEfGA Diagram The block diagram in fig 1 illustrates the transition between two stages of the proposed system. It shows how the embedding and extraction occurs. Though embedding and extraction are two separate stages, the extraction stage occurs only after the embedding stage. Fig1 charts the flow from an input video sample and a watermark sample via obtaining the watermarked video to finally achieving the extracted watermark from the watermarked video. Table 1 shows one of the video frame samples and the watermark sample with its size and format which is going to be tested during implementation. Table 1. Test Video Frame Sample of suzie.avi and Test Watermark Sample of csulogo.png Sample Videos Frame Video Forma t Video Size Sample Watermar k Watermar k Format Watermar k Size avi 176×14 4 png 80×80 A. Proposed Video Watermark Embedding Algorithm (Embedding Stage) In the embedding stage, the flow starts with the input video sample and the watermark sample. There are further sections which are video pre-processing, watermark pre-processing and the embedding process with video post-processing. In video pre-processing, there are sub-sections such as scene change detection, frame color conversion, Nonsubsampled Contourlet Transform, Discrete Wavelet transform and Singular Value Decomposition. In video pre-processing, the tasks are performed with the original input video sample and are related to reducing the computational time and providing high robustness. In watermark pre-processing, there are subsections such as bit plane slicing, modified Arnold transform, and Singular Value Decomposition which are done for the purpose of enhancing the imperceptibility and the payload. Using modified Arnold Transform, a key is obtained which is used to scramble the watermark while embedding and which is later used to descramble while extracting. Table 2. Watermarked Video Frame Sample with its PSNR Value Watermarked Video Frame Imperceptibility (PSNR) 64dB In the embedding process with video post-processing, a watermark embedding process is performed where watermark slices and Eigen Vector of the watermark are embedded using components of transformed domains of the video and all the reversals are carried out so as to obtain the final watermarked video. Table 2 shows that the watermarked video frame sample with its PSNR value. B. Proposed Video Watermark Extraction Algorithm (Extraction Stage) In the extraction stage, similar processes such as video preprocessing and watermark pre-processing are carried out along with watermark detection, extraction and post-processing. However, video-preprocessing is performed in the watermarked video and the original video sample. While performing this task, the additional sub-sections Log Polar Transform and Inverse Log Polar Transform are applied to the videos together with other transform techniques. In watermark detection, extraction and post-processing, the watermark scrambled slices and embedded Eigen Vector are obtained after which processes related to descrambling and reconstructing the watermark are performed. While descrambling, modified Arnold Transform is used. The authentication is done based on comparing Eigen Vector of extracted watermark and extracted Eigen Vector. Table 3 shows the extracted watermark with its NCC value which shows that high robustness of the watermark has been maintained against attacks. Table 3. Extracted watermark with its NCC value Extracted Watermark Robustness (NCC) 0.9973 C. Modified Arnold Transform (mAT) The main limitation of using the traditional Arnold Transformation is that the transform coefficients used are all fixed and once it becomes known that traditional Arnold Transform has been used, then one can somehow descramble the image using these fixed coefficients [15]. Based on the theory provided by [16], modified Arnold Transformation is introduced in our proposed algorithm. The feature of a matrix determinant has been utilized to modify this algorithm, extracting the transform coefficients of Arnold transform in such a way that the determinant of the matrix coefficients is 1. a00* a11 – a10 * a01 = 1 where a00, a11, a10 and a01 are the transform coefficients I. RESULTS AND DISCUSSION This section tests the implementation of the proposed video watermarking algorithm in different attack situations. The tests were carried out using Matlab R2015b on 10 sample videos and 10 color images with watermarks of different sizes and formats. The sample videos and watermarks were collected from a Google database. Since our proposed system has a strong focus on improving the robustness against geometric attacks without affecting the robustness against other attacks, the results and discussions presented below are centered on the comparisons of robustness of the current and proposed system in different attack situations such as several image processing, video processing and geometric attacks. The quality metrics used are peak signal to noise ratio (PSNR) which is used to measure the imperceptibility or quality of the watermarked video frame against the original video frame and a normalized correlation coefficient (NCC) is used to evaluate the correlation of the extracted watermark with the original watermark. Table 4 shows NCC value for one of the sample videos with a sample watermark without any attacks. The table shows that in the absence of attacks, nearly 99% of the watermark can be extracted. According to table 5, comparing the proposed system with the current algorithm, 80% to 99% of the watermark has been extracted from different image processing and temporal attacked watermarked videos from both of them which show that the robustness against image processing attacks and temporal attacks are similar. Table 4. Result of Extracted watermark with NCC value applied in one of the 10 sample watermarked video frames with no attacks Proposed Algorithm (PRVWA-EEfGA) : No attacks Sample Videos Frame Sample Watermark Watermarked Video Frame Extracted Watermark Robustness (NCC) 0.9973 To evaluate performance of our proposed algorithm in terms of robustness, three main geometric attacks, rotation attack, scaling attack and transition attack have bee
n tested. From the table 6 and fig 2, the resulting values of NCC clearly show that our proposed algorithm is able to extract 93% to 99% of the watermark for the rotated watermarked videos. However, the NCC values of the current algorithm ranged from only 20% to 88% for all tested angles except 1800. According to table 7 and fig 3, the resulting values of NCC clearly show that our proposed algorithm is able to extract 97% to 99% of the watermark for the scaled watermarked videos, whereas, for the scaled watermarked video of the current algorithm, the NCC values only ranged from 68% to 84%. Similarly, according to table 8 and fig 3, the resulting values of NCC clearly show that our proposed algorithm is able to extract 94% to 99% of the watermark for the translated watermarked videos, whereas, for the translated watermarked video using the current algorithm, the NCC values only ranged from 38% to 80%. Furthermore, while increasing the robustness of video watermarking algorithm, we have maintained the computational time as low as possible. Table 9 shows the computational time for watermark embedding and extraction for the current and the proposed algorithms which clearly show that only a slight increase in extraction time has occurred which is negligible. The comparison table clearly shows that our proposed algorithm has significantly enhanced robustness against rotation attack for all degrees of rotations, scaling attack and translation attack without affecting other factors. Fig. 2. Comparison of NCC values for different degrees of rotation between current algorithm and proposed algorithm Fig. 3. Comparison of NCC values for scaling attack and transition attack between current algorithm and proposed algorithm II. CONCLUSION In conclusion, the need for digital video watermarking in today’s evolving technological environment is high and development of robust mechanisms is a priority in terms of copyright protection and authentication. A significant number of algorithms have been introduced in order to provide the best possible features required for video watermarking purposes. Our proposed system is based on the current algorithm provided by [3]. The main contribution of our proposed system is through the use of Log Polar Transform and its inverse during extraction only, since the level of watermarked video degradation that occurs does not matter during extraction. Furthermore, this system provides dual security with more secure features than the existing algorithm by using modified Arnold transform and authentication with Eigen Vector. Based on a detailed comparison between the current and the proposed system, the latter has maintained higher imperceptibility, payload, and achieved higher levels of robustness against several types of attacks, particularly against Table 5. Result for robustness against different image processing and temporal attacks for current algorithm and proposed algorithm Sample Video Frame Sample Water mark Robustness for Image Processing Attacks and Temporal Attacks (NCC value) Salt & Pepper Attack (var=0.03,0.01) Gaussian Noise Attack (var=0.1,0.01) Poisson noise Attack Median Filtering Attack Contrast Adjustment Histogram Attack Frame Dropping Attack 21%,58%,96% Frame Swappi ng Attack 25% Current Algorithm Agilandeeswari, L., & Ganesan, K. (2015) 0.8438 0.9557 0.7025 0.7497 0.9822 0.9852 0.9993 0.9678 0.9525 0.9368 0.8200 0.9622 Proposed Algorithm (PRVWA-EEfGA) 0.9972 0.9973 0.9968 0.9974 0.9974 0.9974 0.9975 0.9976 0.9405 0.9259 0.7999 0.9622 Table 6. Result for Robustness Against Different Rotation Attacks for Current Algorithm and Proposed Algorithm Angle of Rotation (In Degree) Sample Rotated Video Frames Current Algorithm Agilandeeswari, L., & Ganesan, K. (2015) Proposed Algorithm (PRVWA-EEfGA) Extracted Watermark Robustness (NCC) Extracted Watermark Robustnes s (NCC) 1 0.8635 0.9972 5 0.3013 0.9968 10 0.3249 0.9970 45 0.4057 0.9971 135 0.5310 0.9967 180 0.9999 0.9968 main geometric attacks than the current algorithm. Furthermore, the proposed system has achieved this without affecting computational time to any marked degree and also enhanced the security features by modifying the traditional Arnold Transformation. There are plans to conduct further research to reduce computational time while maintaining all the acquired features. Table 7. Result for Robustness Against Scaling Attack for Current Algorithm and Proposed Algorithm: (Resized by Width-100, Height-100 and then resized to its original size) Current Algorithm Agilandeeswari, L., & Ganesan, K. (2015) Proposed Algorithm (PRVWA-EEfGA) Sample Videos Frame Extracted Waterma rk Robustne ss (NCC) Sample Videos Frame Extracted Waterma rk Robustne ss (NCC) 0.6815 0.9973 Table 8. Result for Robustness Against Transition Attack for Current Algorithm and Proposed Algorithm Transition Value = x=10.3, y=-10.1 Current Algorithm Agilandeeswari, L., & Ganesan, K. (2015) Proposed Algorithm (PRVWAEEfGA) Sample Transite d Video Frames Extracted Watermar k Robustnes s (NCC) Sample Transite d Video Frames Extracted Watermar k Robustnes s (NCC) 0.7266 0.9977 Table 9. Result for Computational time taken by each frame for embedding and extraction by Current Algorithm and Proposed Algorithm Sample Video Frame Sample Watermark Embedding and Extraction Time Current Algorithm Agilandeeswari, L., & Ganesan, K. (2015) Proposed Algorithm (PRVWA-EEfGA) 0.07s 0.06s 0.07s 0.093s REFERENCES [1] S. Bhattacharya, T. Chattopadhyay and A. Pal, “A Survey on Different Video Watermarking Techniques and Comparative Analysis with Reference to H.264/AVC,” ISCE ’06. 2006 IEEE Tenth International Symposium on Consumer Electronics, pp. 1-6, 2006. [2] R. Bala, “A Brief Survey on Robust Video Watermarking Techniques,” The International Journal Of Engineering And Science, vol. 4, no. 2, pp. 41- 45, 2015. [3] L. Agilandeeswari and . K. Ganesan, “A robust color video watermarking scheme based on hybrid embedding techniques,” Multimedia Tools and Applications, pp. 1-36, August 2015. [4] N. K. Dubey and S. Kumar, “An Effective Approach of Distortion- Resistant Video Watermarking for Piracy Deterrence,” International Journal of Security and Its Applications, vol. 9, no. 1, pp. 283-294, 2015. [5] H. Chen and Y. Zhu, “A robust video watermarking algorithm based on singular value decomposition and slope-based embedding technique,” Multimedia Tools and Applications, vol. 71, no. 3, pp. 991-1012, 2012. [6] D. Li, D. Jiang and J. W. Kim, “Robust Adaptive Video Watermarking Algorithm based on Dual Transform Domains,” International Journal of Multimedia and Ubiquitous Engineering, vol. 9, no. 1, pp. 391-402, 2014. [7] S. Youssef, A. Elfarag and N. Ghatwary, “Adaptive video watermarking integrating a fuzzy wavelet-based human visual system perceptual model,” Multimedia Tools and Applications, vol. 73, no. 3, pp. 1545-1573 , 2014. [8] J. Xuemei, L. Quan and W. Qiaoyan, “A new video watermarking algorithm based on shot segmentation and block classification,” Multimedia Tools and Applications, vol. 62, no. 3, pp. 545-560, 2013. [9] C. Narasimhulu and K. S. Prasad, “A Robust Watermarking Technique based on Nonsubsampled Contourlet Transform and SVD,” International Journal of Computer Applications, vol. 16, no. 8, pp. 27-36, 2011. [10] L. Gong, X. Liu , F. Zheng and N. Zhou, “Flexible multiple-image encryption algorithm based on log-polar transform and double random phase encoding technique,” Journal of Modern Optics, vol. 60, no. 13, pp. 1074- 1082, 2013. [11] R. Yang, X. Kang and J. Huang, “Robust Audio Watermarking Based on Log-Polar Frequency Index,” Digital Watermarking, vol. 5450, pp. 124-138, 2009. [12] J. Ouyang, . G. Coatrieux , B. Chen and H. Shu, “Color image watermarking based on quaternion Fourier transform and improved uniform log-polar mapping,” Computers and Electrical Engineering, vol. 46, pp. 419- 432, 2015. [13] L. Wang, H. Ling, F. Zou and Z. Lu, “Real-Time Compressed- Domain Video Watermarking Resistance to Geometric Distortions,” MultiMedia, vol. 19, no. 1, pp. 70 – 79, 2012. [14] D. Li, L. Qiao and J. Kim, “A video zero-watermarking algorithm based on L
PM,” Multimedia Tools and Applications, pp. 1-14, 2015. [15] G. V. Artist and M. K. Porwal, “Dual Layer Image Scrambling Method Using Improved Arnold Transform,” American International Journal of Research in Science, Technology, Engineering & Mathematics, vol. 9, no. 3, pp. 258-264, 2015. [16] Z. Shang, H. Ren and . J. Zhang, “A Block Location Scrambling Algorithm of Digital Image Based on Arnold Transformation,” International Conference for Young Computer Scientists, pp. 2942-2947, 2008. 7 Table 6. Result for Robustness Against Different Rotation Attacks for Current Algorithm and Proposed Algorithm Angle of Rotatio n (In Degree) Sample Rotated Video Frames Current Algorithm Agilandeeswari, L., & Ganesan, K. (2015) Proposed Algorithm (PRVWA-EEfGA) Extracted Watermar k Robustnes s (NCC) Extracted Watermar k Robustnes s (NCC) 1 0.8635 0.9972 5 0.3013 0.9968 10 0.3249 0.9970 45 0.4057 0.9971 135 0.5310 0.9967 180 0.9999 0.9968 8

Leave a Reply

Your email address will not be published. Required fields are marked *