2017 Artificial Intelligence and Robotics (IRANOPEN) Classification of Watermarking Methods Based on Watermarking Approaches Mahsa Boreiry Mohammad-Reza Keyvanpour Department of Computer Engineering Islamic Azad University (Qazvin Branch) Qazvin, Iran Mahsa.Boreiry@gmail.com Department of Computer Engineering Alzahra University Tehran, Iran keyvanpour@Alzahra.ac.ir which may be embedded several times. Even though these are often lost during the attack, the survival of a small watermark can be accounted as a success. Despite of its simplicity, there would be some problems. For example, this method is resistant against cropping attacks but a little noise, waste or compression can breakdown the watermark. Fig. 1 illustrated the sample of embedding and extraction of watermark with LSB method. Abstract— Video watermarking is the approach of adding insensible data to video content in order to protect the owner’s right. One of the main important issues in watermarking operation is the watermarking robustness. The basic requirement in watermarking is resisting in front of distortion and initial attacks, which is commonly examined based on the data processing standards. Many methods have been proposed in the field of video watermarking that are resistant to certain attacks, therefore some attacks can break the watermark. So, the correct identification of methods and knowing the strength and weakness of each method lead to propose appropriate solution in order to reduce the effects of attacks with presenting secure algorithm in video watermarking. Keywords— Watermark, Video Watermarking, Watermarking methods, Watermarking approaches. I. INTRODUCTION a) Embeding watermark b) Watermark extraction Fig. 1 Embedding and extraction of watermark with LSB method Generally watermarking methods can be classified based on embedding domain to spatial methods (pixel domain approach) and spectral domain (transformation domain approach). In this study the concentration is focused on spatial embedding methods. In the following, each category will be evaluated beside the relative examples. In this article, the watermarking methods will be evaluated and categorized based on the watermarking approach, while the focus is paid to investigate some methods in each category with declaring the advantages and disadvantages of each case[1]. B. Spread Spectrum Method In the basic algorithm each bit of watermark aj , aj ∈[1] is spread over a large number of chips and modulated by a binary pseudo-noise sequence pj , pj ∈[1]. By the means of high-pass filtering with correlation based method, the watermark can be retrieved[2]. Diagram of this method shown in Fig. 2. C. 2D Spread Spectrum A 2D spread spectrum method is utilized to monitor video data transmitted over different broadcast links for video watermarking (just another watermarking system, JAWS)[3]. II. EMBEDING WATERMARK DOMAIN USING SPATIAL METHODS In the spatial domain watermarking techniques, the watermark embedded in the host image/video by changing its pixel values directly. The spatial domain watermarking can hardly survive under the attacks such as lossy compression and low-pass filter. The main goal of such methods which are based on pixels are conceptually simple with poor computational complexities. As a result, most of these methods are applicable in real-time applications. In contrast, the watermark optimization is difficult using only spatial analysis techniques[1]. Fig. 2 Spread Spectrum diagram D. CDMA method In CDMA technique, one of the four least significant bitplanes will be replaced by watermark planes. The random periodic quaternary sequence is used to select the bit-planes that should be replaced. The 1D spread spectrum methodology is A. LSB method The LSB method is the most frequently used method for embedding watermark in this field. Due to provide extraordinary channel capacity the whole cover for a small object is used, 978-1-5386-2862-1/17/$31.00 ©2017 IEEE 73 employed to generate the watermark plane. This achievement is applicable for the case in which the accumulation of signals overlap in time and frequency domain and remain separate. The essence of CMDA is hiding the sequence of messages with a covering signal such as noise[4]. Fig. 3 illustrated a CDMA transmitter and receiver. A. DCT method The first efficient layout of watermarking was introduced by Knoch et al. In this method, the image was initially divided into blocks of size 8×8, from 12 preset pairs, 2 pairs of the frequency average coefficients are selected to transform. Then, Brush and Pietas have developed a method to change the coefficient of DCT and improve the restrictions of the selected block. After dividing the images into blocks with size 8×8 the main blocks are chosen based on the classification decisions Gaussian of Network (Fig. 4). Then, using the DCT’s linear constraints the middle range of frequency coefficients change. In transform domain methods, the host signals is transformed in various domains while the watermark embedded in chosen coefficients[5]. The commonly used methodologies are Discrete Cosine Transformation (DCT) and Discrete Wavelet Transformation (DWT)[2]. Block diagram of DCT watermarking is shown in Fig. 5. Fig. 3 CDMA Transmitter and receiver E. Region based energy modification (RBEM) method In this method for embedding data the average energy or luminance intensities will be manipulate in sub-region of each frame. This method embeds 1 bit per 8 × 8 block to achieve high data capacity, besides the error control method is utilized to ensure the resistance. A summary of investigation in video watermarking methods in the spatial domain are illustrated in table 1. Fig. 4 Generating frequency coefficients of image by DCT III. EMBEDDING WATERMARK DOMAIN IN FREQUENCY DOMAIN In compare with watermarking with spatial domain, watermarking in frequency domain which uses general methods and image compression standards is stronger and more consistent. Though, watermarking in frequency domain attracted more attention. In this category, the watermark will be embedded by changing the coefficient of watermark, after applying the frequency transformation on data host. The image frequency transformation will be categorized into Discrete Fourier Transformation (DFT) and Discrete Cosine Transform (DCT). Fig. 5 Block diagram of DCT Watermarking TABLE 1: INVESTIGATION OF VIDEO WATERMARKING METHODS IN SPATIAL DOMAIN advantages Disadvantages In this method, by using energetic signal propagation a high resistance can be achieved. The blind watermarking technique in order to embed the watermark don’t use the host signals. Do not specifically protect the value of DC blocks. LSB Fail in facing with cropping attacks, compression, low-pass filter. Robustness restriction, capacity limitation in data storage and low resistance Spread Spectrum Resistance to geometric attacks such as removal of inner distance, scaling, rotation, simplicity and lack of computational complexity and conceptual clarity This method change pixel value for embedding watermarking. Each bit of watermark aj is spread over several chips and modulated by a binary pseudo-noise sequence. The watermark is embedded in a vector form. Data recovery is done by the means of highpass filter. A 2D spread spectrum Main idea A watermark pattern S × S is created in the beginning while this pattern is embedded alternatively. Around some points will be chosen to be fixed. After generating the watermark frames using the host masks the spatial mask will be applied on them. A little calculation will be used in detection algorithm. CDMA Method In CDMA technique one of the four least significant bit-planes will be replaced by watermark planes. The random periodic quaternary sequence is used to select the bit-planes that should be replaced. This method has more data capacity for watermark. 74 This technique in blind watermarking technique that embeds a watermark signal without using host signals will fail. Don’t specifically protect from the value of DC blocks. Fail in facing with statistical attacks (average frames) TABLE 3: COMPARISON BETWEEN WATERMARKING METHODS B. TDC method As one of the first transformed domain video watermarking methods (TDC) the work done by Cox et al. in [6] can be mentioned. This study is concentrated on the importance of embedding the watermark into perceptually significant components in order to increase robustness against signal processing and lossy compression techniques. A non-blind approach is utilized for watermark detection. The detection is performed by the original and test frame in the DCT domain, besides it correlates the difference vector with the expected watermark pattern. C. PW method A perceptual watermarking (PW) method which model masking properties of the HVS, analyze video sequence or frames using the mentioned models leads to embed watermark in the optimal way. The main properties of the HVS named as: frequency sensitivity, luminance sensitivity, contrast masking, edge masking and temporal masking, which are exploited by video watermarking techniques[7, 8]. Spatial domain Spectral domain Computational complexity low high Computation time low high Resistance fragile resistant Capacity high low Application Authentication Copy control Due to evaluate more accurate and compare the approaches and methods, summary of mentioned methods are listed in table 4,5. Table 4 reviews the overall watermarking methods based on the transformation domain, while table 5 compares the methods based on their configurations. A summary of investigation of video watermarking methods in the frequency domain are depicted in table 2. TABLE 4: CLASSIFICATION OF METHODS BASED ON WATERMARK TABLE 2: INVESTIGATION VIDEO WATERMAKING METHODS IN FREQUENCY EMBEDDING TECHNIQUE DOMAIN The watermark of length n was populated from a standard normal distribution apart from a binary PN sequence in order to enhance robustness. A perceptual watermarking (PW) method which models masking properties of the HVS, analyzes video sequence or frames using the mentioned models leads to embed watermark in the optimal way. Showing more resistance against attacks in compare with spatial domain. Domain-specific features (such as HVS) can be easily enforced. method disadvantages Complicated calculations. Highfrequency components tend to remove in compression level. Pixel domain approach advantages Relatively high time complexity More resistant to signal processing techniques and lossy compression. Main idea In this approach, data will be stored in source media by varying the pixel values using the computational methods Transformation domain approach Main idea First, the host video transforms by using frequency domain methods, then the transformation coefficients domain change for embedding watermark data. Finally, the inverse transformation is used to obtain the image of watermarked video. In this approach, frequency transformation will be applied on host, for example Discrete Fourier Transformation (DFT) and Discrete Cosine Transform (DCT) Compression domain approach PW TDC DCT method In this approach, the computational tool is utilized for compression in watermarking. For example JPEG2000, MPEG,H.264 and … compressions. advantages disadvantages simplicity Fail in blind watermarking More resistance against general image processing such as low-pass filter, adjust brightness, contrast and blurring the image Reduce the complexity of Computational complexity, low resistance against some process as rotation, cropping and changing size Poor resistance and reliability IV. CONCLUSION The main goal of this manuscript is to investigate the video watermarking methods with declaring the strengths and weaknesses of each method. Of course, each case besides the weakness is suitable for certain applications in watermarking. And also, besides the mentioned methods, combining the basic approaches in this field will lead to increase the efficiency of proposed algorithm. By considering the expected application The comparison of watermarking methods using spatial domain and spectral domain are presented in table 3. 75 and investigating event vulnerability, appropriate solution can be represented due to prepare minimum security requirements. [4] B. G. Mobasseri, "Exploring CDMA for watermarking of digital video," in Electronic Imaging'99, 1999, pp. 96-102. [5] S. Patel, A. Katharotiya, and M. Goyani, "A Survey on Digital Video Watermarking," International Journal Comp. Tech. Appl, vol. 2, pp. 30153018, 2011. [6] I. J. Cox, J. Kilian, F. T. Leighton, and T. Shamoon, "Secure spread spectrum watermarking for multimedia," IEEE transactions on image processing, vol. 6, pp. 1673-1687, 1997. [7] R. B. Wolfgang, C. I. Podilchuk, and E. J. Delp, "Perceptual watermarks for digital images and video," Proceedings of the IEEE, vol. 87, pp. 11081126, 1999. [8] M. Reid, R. J. Millar, and N. D. Black, "Second-generation image coding: an overview," ACM Computing Surveys (CSUR), vol. 29, pp. 3-29, 1997. REFERENCES [1] T. Jayamalar and V. Radha, "Survey on digital video watermarking techniques and attacks on watermarks," International Journal of Engineering Science and Technology, vol. 2, pp. 6963-6967, 2010. [2] S. Bhattacharya, T. Chattopadhyay, and A. Pal, "A survey on different video watermarking techniques and comparative analysis with reference to H. 264/AVC," in 2006 IEEE International Symposium on Consumer Electronics, 2006, pp. 1-6. [3] F. H. Hartung and B. Girod, "Digital watermarking of raw and compressed video," in Advanced Imaging and Network Technologies, 1996, pp. 205213. TABLE 5: COMPARISION BETWEEN METHODS BASED ON CONFIGURATIONS approach Pixel domain approach Compression domain approach Transformation domain approach method SS JAWS resistance acceptable acceptable reliability acceptable acceptable invisibility good good applicability good good Time complexity good good CR good good good poor poor CDMA acceptable acceptable good acceptable acceptable RBEM VLC acceptable poor acceptable poor good good acceptable good acceptable good H.264 TDC good good good good good good acceptable acceptable acceptable acceptable PW acceptable acceptable good acceptable good DCT good good good good acceptable 76