Author Guidelines for 8

6 downloads 0 Views 273KB Size Report
DIGITAL VIDEO WATERMARKING BASED ON YCBCR COLOUR SCHEME. N. Alharbi, C. Schofield, R.S. Thammanna. Department of Computer Science, the ...
DIGITAL VIDEO WATERMARKING BASED ON YCBCR COLOUR SCHEME N. Alharbi, C. Schofield, R.S. Thammanna Department of Computer Science, the University of Sheffield, Regent Court, 211 Portobello, Sheffield, S1 4DP. United Kingdom. ABSTRACT Video watermarking is well known as the processes of embedding copyright information in video bit streams. It had been proposed in recent years to solve the problem of illegal manipulation and distribution of digital video. In our work, we have proposed two blind video watermarkings based on the YCbCr colour space algorithm and RGB colour space algorithm, due to its robustness and imperceptibility. This watermark can be detected without the original video in both the algorithm based on the RGB colour space and another one based on the YCbCr colour space. The performance of the two proposed algorithms is evaluated with respect to imperceptibility and robustness. The performance of the algorithm YCbCr colour space is better than the algorithm RGB colour space. Index Terms— Digital video watermarking, copyright protection, YCBCR colour space, imperceptible and robust watermarking 1. INTRODUCTION Video Watermarking is a relatively new technology, and has arisen mainly to combat the illegal distribution of digital video over media such as CD/DVD/Blu-ray and, more predominantly, the internet. The aim of any video watermarking scheme is to embed extra information into the video in such a way that it is perceptually undetectable, while still holding enough information to be able to extract the original watermark from the resultant video. This is to allow the owner of such videos to establish ownership through showing the presence of the watermark, which contains identifiable information such as a company logo. Many research projects have taken place that have brought about the existence of many potential algorithmic solutions to this problem, with one of the most successful of these being Digital Watermarking [8][9]. Digital Watermarking involves taking some medium, whether text, audio, still images or video, and embedding it with extra information that is very hard (ideally impossible) to remove without authorization [6] or harming the value of the data, as well as being perceptually unnoticeable. This means its presence should not affect the original content of the host data. Every watermarking system consists at least of

two different parts: watermark embedding unit and watermark detection and extraction unit; for each part there are multiple alternatives in existence [7]. The main aim of this study has been to implement a blind system which enables the protection of a video from copyright infringement, illegal distribution and manipulations, while not requiring the original video on the receiver side to verify the validity of permissions. We also aimed to implement a watermarking extraction procedure that will take a watermarked video and extract the original watermark, ideally in its original form. This may equate to the owner applying the watermark extraction technique to a video they claim to own in order to prove that they do. As a final objective, we hoped to test our implementation with various types of video and still images. This had the extended objective of performing simple tests against still images with different levels of colour and resolutions, before extending our testing to full video scenes. At the end of the implementation stage, the resultant algorithm was also checked for robustness by simulating a range of attacks on the watermarked video with the aim of circumventing the watermarking scheme and reducing the ability to extract the watermark. These robustness checks included resizing the video, swapping or removing frames from the video, and so on [2][4][5]. 2. WATERMARKING TECHNIQUES Watermarking techniques can be divided into two main categories. The first category describes correlation-based methods; the second category comprises the non-correlationbased techniques [3]. Algorithms of the first category embed digital watermarks by adding pseudo-random noise to the image components, which are detected by correlating the image noise with the components of the image. The second category can be subdivided into least significant bit and geometrical relation techniques. The most commonly used watermarking methods are based on correlation techniques. Researchers tend to investigate video watermarking techniques that are robust and invisible. Spatial Domain Watermarking: In the spatial domain technique the watermark embedding is achieved by directly modifying the pixel values of the host image [1]. The most commonly used method in the spatial domain technique is the least significant bit (LSB). In this approach, the least

significant bit (LSB) of each pixel in the host image is modified to embed the secret message. This method can only resist some attacks. Frequency Domain Watermarking: In the transform domain technique, the host image is first converted into frequency domain by a transformation method such as discrete cosine transforms (DCT), discrete Fourier transforms (DFT) or discrete wavelet transforms (DWT). Then, transform domain coefficients are modified by the watermark. The inverse transform is finally applied in order to obtain the watermarked image. The transformation domain methods are more robust against attacks than spatial domain methods [1]. Spread Spectrum Watermarking: Spread spectrum techniques used in digital watermarking are borrowed from the communications field. The basic idea of spread spectrum is to spread the data across a large frequency band. Spread spectrum is a military technology designed to handle interferences and disturbances. Spread spectrum communication is used in radar, navigation, and communication applications. The information is weaved into the source material using a secret key or an embedding procedure [3]. 3. THE RGB AND YCBCR COLOUR SPACES RGB (Red, Green and Blue) is an additive colour model in which varying proportions of red, green and blue are added together to create different colours. The RGB colour scheme is primarily associated with the display of colour images on electronic devices such as televisions or computer monitors, though it also appears in photography. A colour in the RGB colour space is represented numerically as a triplet, where each value of the triplet specifies the amount of red, green or blue, respectively. YCbCr is a colour scheme used primarily in digital photography, and while it is not an absolute colour space, it is a way of encoding RGB information. Here, Y is the luma component, while Cb and Cr are the bluedifference and red-difference chroma components, respectively. For example, in image encoding formats such as JPEG, image data is stored internally in the YCbCr colour space, allowing lossy sub-sampling to be performed on the chroma channels, reducing the overall file size. YCbCr is also used in television broadcasts, in which images in RGB are encoded using a lossless encoding technique before broadcast, and then decoded back to RGB on receipt at the destination. 4. PROPOSED ALGORITHM As part of our study, we have arrived at an algorithm that is proposed in this section for the embedding and extracting of an image watermark in a target video. It consists of the following principle procedures: 1. Pre-processing: This stage is where a target video is split into frames, followed by skin detection being carried out on these frames. This results in a set of skin-

2.

3.

4.

detected frames that contain yellow pixels (where the algorithm has detected skin) or black pixels. Choosing candidate watermark target frames: This stage consists of examining the frames outputted in the previous step to establish the percentage coverage of yellow pixels of each frame. Those frames that contain over a certain threshold of yellow pixels (i.e. those images containing a lot of skin, which we deemed as being high-value content frames) are added to a list of candidate watermark target frames. Embedding: This stage takes a subset of the frames derived in the previous step and applies the embedding algorithm on each of these frames. Once this has been done, the watermarked video is constructed through a combination of the original untouched frames and the watermarked frames. Extraction: In the extraction step, the embedding steps are essentially repeated except the values of the watermark image are gradually reconstructed by reading from the watermarked frames, arriving at a reconstructed watermark image.

4.1. Pre-processing of the target video As stated above, this stage involves splitting the video into frames and performing skin detection on those frames. It is carried out via the following algorithm: 1. 2.

3.

4. 5.

Choose a target video on which the watermark is to be embedded. Split the video into individual frames, outputting the number of frames per second multiplied by the number of seconds (for example, a 10 second video at 15 frames per second should output 150 frames). For each of these frames, perform the skin detection algorithm on it as below: a. Read the image. b. Convert the image to the YCbCr colour scheme. c. For each pixel value, analyze its colour, and if the colour closely matches the desired skin colour, output a yellow pixel, otherwise output a black pixel. d. Once the whole frame has been processed, a new frame consisting of yellow and black is created which is saved to disk. Repeat this procedure for every frame. At this stage, the set of skin-detected images has been created.

4.2. Choosing candidate watermark target frames The next step involves counting the number of yellow pixels that exist in each skin-detected frame and using the total number of pixels in the whole frame along with this value to calculate the percentage of yellow pixels in the image. This step is carried out via the following procedure: 1. Read a frame from the set of skin-detected frames.

2. 3. 4.

5.

6.

Define a counter with initial value 0. For each pixel in this frame, if it is yellow, add 1 to the counter, otherwise ignore. Once the frame has been processed completely, divide the number of yellow pixels by the total number of pixels to arrive at a percentage of yellow pixels. If this value is over some threshold (for example 10%), then add this frame to the list of candidate watermark target frames. Repeat this procedure for every skin-detected frame, arriving at a list of candidate watermark target frames.

1.

2.

3. 4. 5.

4.3. Embedding the watermark Once the set of skin-detected frames has been used to create a list of candidate watermark target frames, the next step in the algorithm is to embed the watermark image. This step is carried out via the following procedure: 1.

Take a subset of the candidate target frames and choose a watermark image. 2. Take the first frame to be embedded, and apply the watermark as follows: a. Resize the target frame to 256x256 and watermark frame to 16x16, and convert both into the YCbCr colour scheme. b. Partition the target frame into blocks of 8x8 pixels and choose 3 within the block. c. Take the first pixel of the watermark image and separate the Y, Cb, and Cr components of the pixel value. d. Embed the Y, Cb, and Cr components of the first pixel of the watermark image into the three chosen pixels in the first block of the watermark target frame. e. Move to the next block and apply the same process with the Y, Cb and Cr components of the second pixel of the watermark image and continue this procedure for every block in the target frame. 3. Repeat this procedure for all frames to be watermarked. 4. Once the watermarking is complete, construct the watermarked video by taking the unaffected original frames along with the watermarked frames, arranging these into the correct order and saving them to disk. 4.4. Extracting the watermark Once the watermarked video has been created, the extraction procedure can be carried out upon it, which will yield the original watermark image. This procedure is as follows:

6.

Split the video into frames and perform skin detection on those frames, creating the list of candidate frames as above, only this time these will be candidate extraction frames. Choose the frames to be extracted using the same choice method as above, so that the same frames are used for extraction as were used for embedding the watermark. Resize the frame to 256x256 and convert to YCbCr as above. Partition the frame into blocks of 8x8 as above. Take the first block and get the three pixel component values from it by reading the same pixels within this block as were used in embedding. Use these component values to derive the overall pixel value of the first pixel in the watermark image. Repeat this procedure for all blocks to arrive at the watermark image and output the extracted watermark to disk.

Fig. 1: YCBCR watermark embedding and extraction procedures

5. EXPERIMENTAL RESULTS AND PERFORMANCE EVALUATION In order to evaluate the overall performance of the proposed algorithm, we performed a series of tests including the measurement of imperceptibility performance. In addition to this, we also tested the robustness of the proposed algorithm by performing a series of attacks on the watermarked video. This was with the aim of simulating the effect that somebody trying to circumvent the watermark would have, and to determine whether our algorithm was able to withstand such attacks. 5.1. Imperceptibility performance Imperceptibility means that the perceived quality of the video clip should not be distorted by the presence of the watermark. As a measure of the quality of a watermarked video, the peak signal-to-noise ratio (PSNR) was calculated. This function displays the PSNR between two images. The average PSNR for all frames was 41.63, and this high value proves the imperceptibility of the proposed algorithm. 5.2. Compression The actual compression codec of the watermarked video is changed from mpeg to avi. Our algorithm was evaluated

against the baseline algorithm. The results show that the watermark can still be extracted from all the videos that were watermarked using the proposed algorithm, against 85.7% of the videos watermarked using baseline algorithm. 5.3. Changing from colour to greyscale This is when the video is changed from a colour video to a greyscale video. Again, doing this shouldn't affect our ability to extract the watermark. 5.4. Resizing attack The nature of this attack is such that when the frames of a video are resized, data may be lost if the compression codec used is lossy, to the point where resizing the frames causes minimal visual damage but sufficient low-level damage such that the ability to extract the watermark from the video is lost or reduced. We tried both enlarging and shrinking the watermarked video prior to extraction. Presently, the watermarked video produced by each approach had dimensions 256 x 256. We conducted prior resizing on each of the watermarked videos by treating each frame as a JPEG image which was then resized before the frames were recombined into a video, changing the dimensions to 64 x 64 and 512 x 512 in each case, before trying to extract the watermark using each extraction approach. The results after conducting these tests are shown in Figure 2 below, where ―Pass‖ means the watermark was still extractable. Results show that for enlarging, the baseline performance is 95.24%, which is improved to 100% when the proposed algorithm is used, and for shrinking, the baseline performance is 42.86%, improved to 61.90% when the proposed algorithm is used.

watermarking. we also concentrated on minimised noise in the watermarked video. We proposed two algorithms in our research study, one based on the RGB and the other based on the YCbCr. These two are robust with different possible dataset. Using the above experimental results, the YCbCr performed better than the RGB. This is due to the fact that after embedding the watermark into the video, using the RGB algorithm produces noise within the video. 6. CONCLUSION This study has explored the relatively new field of Video Watermarking. The study started with the acquisition of basic knowledge of the field and existing work, and progressed into the full development of a new and innovative method of solving the problem of adding information into a video in a way that is not perceivable by humans. The algorithm could be extended to be robust against further forms of attack, by carrying out extra attacks (such as cropping) and devising methods of protecting the integrity of the watermark image in such cases. 7. REFERENCES [1] Dittmann, J., M. Stabenau, and R. Steinmetz (1998). Robust MPEG video watermarking technologies. In Proceedings of the sixth ACM international conference on Multimedia, pp. 71-80. ACM. [2] Fridrich, J. (1998, November 4–6). Applications of data hiding in digital images. Tutorial of the ISPACS ’98 Conference, Melbourne, Australia. [3] Hanjalic, A., Langelaar, G.C., van Roosmalen, P.M.G., Biemond, J., & Langendijk, R.L. (2000). Image and video databases: Restoration, watermarking and retrieval. Amsterdam. [4] Heileman, G.L., Pizano, C.E., & Abdallah, C.T. (1999, September). Performance measures for image watermarking schemes. Proceedings of the Fifth Baiona Workshop on Emerging Technologies in Telecommunications, Baiona, Spain. Retrieved February 14, 2004.

Fig. 2: result of testing YCBCR and baseline algorithms against resizing attack

5.5. Frame Swapping Attack The nature of this attack is such that the frames in the watermarked video are shuffled to cause minimal damage to the usefulness of the video while causing sufficient damage to remove the ability to extract the watermark image. We have developed an algorithm that shuffles the frames in the watermarked videos and results provided to the extraction algorithms to see if the watermark can still be extracted. The results of performing this attack on the watermarked videos are show that the baseline performance is 85.71%, improved to 95.24% when the proposed algorithm is used. 5.6 Discussion Digital watermarking is the process of embedding information into a digital signal in a way that is difficult to remove. Since we are primarily interesting in invisible

[5] Kutter, M., & Hartung, F. (2000). Introduction to watermarking techniques. In S. Katzenbeisser & F.A.P. Petitcolas (Eds.), Information hiding techniques for steganography and digital watermarking. Boston. [6] Mohan, B., S. Srinivaskumar, and B. Chatterji (2008). A Robus digital image watermarking scheme using singular value decomposition (SVD), dither quantization and edge detection. ICGST-GVIP J 8, 17-23. [7] Perez-Gonzalez, F. and J. Hernandez (2002). A tutorial on digital watermarking. In Security Technology, 1999. Proceedings. IEEE 33rd Annual 1999 International Carnahan Conference on, pp. 286-292. IEEE. [8] Rajab, L, Al-Khatib, T, Al-haj, A (2009). Video watermarking algorithms using the SVD transform. [9] Rajab, L., T. Al-Khatib, and A. Al-Haj (2009). Video Watermarking Algorithms Using the SVD Transform. European Journal of Scientific Research 30(3), 389-401.