Abstractin this paper, we propose a digital watermarking
method for stereoscopic video frames. A stereoscopic video frame
is stored as two images and combined by software while displayed. These two stored images are with high similarity. In the circumstances, it should be taken into consideration to resist coalition attacks while developing a watermarking scheme for stereoscopic videos. As shown in experimental results, the watermark embedded with the proposed scheme is still detectable after common image processing operations such as JPEG compression, Gaussian filtering and sharpening. In the future, we will develop stereoscopic video watermarking methods based on the proposed method. I. INTRODUCTION Because of the popularity of on-line video sharing sites, besides p2p and on-line file storage, pirated versions of video are also distributed by these web sites, now. Up to the present, an on-line video sharing site has removed these pirated video according to the reports from users. However, many users may watch and download the video before the website removes it. If a pirated version of video can be detected when it is being uploaded, the economy loss caused by the pirated version can be future reduced. Thus, we proposed to embed a watermark into the video content [2-5] before it is published. When a video clip is uploaded to the on-line video sharing site, the website will judge if a watermark is present in the video to confirm its copyright status. If this video clip should not be shared on the Internet, the web site can then refuse the file transfer. This detection operation may take a significant amount of computational cost due to the large amount of video content. Therefore, the watermark embedding operation can be time consuming, but the detection method must be efficient and effective. In this paper, we proposed a digital watermarking approach [2] for stereoscopic video frames (i.e., 3D images). With slight modification, this method could also be applied to stereoscopic video contents. To increase the efficiency of the watermark detection operation, the proposed detection method can increasingly extract watermark data from a portion of a stereoscopic video frame while the frame is receiving. Therefore, whether the video frame contain a watermark can be confirmed according to a part of the image data. Then, the website may decide to refuse a pirated video content to save storage and bandwidth.
II. THE PROPOSED ALGORITHM A stereoscopic video frame contains two images, which are taken in different positions at the same time. Two cameras placed side-by-side, approximately 6 cm apart, are used to capture these images from slightly different positions for the left and right eyes. An example of stereoscopic video frames is shown in Figure 1. Since a large amount of contents of these two images are similar, the coalition attack must be taken into consideration when a watermark is embedded. In other words, an attacker may make use of the similarity of these two images to remove the embedded watermark. Therefore, the proposed watermarking method will firstly identify the similar portions of the two images containing in a stereoscopic video frame. Then, identical watermark data will be embedded into the same positions of the similar portions to resist coalition attack.
The proposed watermark embedding method involves eight steps. These steps are described as follows. 1. A stereoscopic video frame is divided into two images. The image on the left side is referred to as image A and the other as image B. Supposed that dimensions of image A and image B are w u h. 2. Then, mean squared error (MSE) between the right portion of image A and the left portion of image B with dimensions i u h, 1d i dw, is calculated. Assumed that minimal MSE value is obtained when i = W s . 3. As shown in figure 2, image A and B are divided into sub-image a, b, c and d. Sub-image b and c are similar in the metric of MSE. Thus, sub-image b and c should contain identical watermark data. 4. Apply discrete wavelet transform to image A and B,
A Robust Watermarking Scheme for Stereoscopic Video Frames Yueh-Hong Chen Dept of Comp. Sci. Info. Eng. Far East University Tainan, Taiwan, R.O.C. yuehhong@gmail.com Hsiang-Cheh Huang Dept of Electrical Eng National Univ. of Kaohsiung Kaohsiung, Taiwan, R.O.C. hch.nuk@gmail.com Figure 1. An example of the stereoscopic video frames. This image is stored with side-by-side format. 978-1-4673-6199-6/13/$31.00 2013 IEEE 2013 IEEE 17th International Symposium on Consumer Electronics (ISCE) 295
respectively. Then, method proposed in [1] is used to embed an identical binary watermark into image A and B. 5. Before each watermark bit is embedded, a random number is generated to decide the watermark bit will be embedded into sub-image b and c, or sub-image a and d. 6. Assumed that sub-image b and c are chosen. Then, wavelet coefficients corresponding to sub-image b and c are random selected. These coefficients are then modified with the method proposed in [1].to embed the watermark bit. Since the same random number is used to select one coefficient from sib-image b and one from sub-image c, it is clear that they are at the same position in sub-image b and c. 7. If sub-image a and d are chosen, coefficients at different positions in sub-image a and d are randomly chosen to embed the watermark bit. 8. Apply inverse wavelet transform on image A and B and combine them into a watermarked stereoscopic video frame. Keep W s and the seed of the random number generator as our secret key. As described above, each pair of coefficients from sub- image b and c are at the same position; moreover, the method proposed in [1] uses the relationship rather than the value of coefficients to hide watermark data. Therefore, the watermark may not be removed even an attacker exchange or average similar pixels in the left image and right image of a watermarked stereoscopic video frame. III. EXPERIMENTAL RESULTS Up to now, there has been no standard test image in stereoscopic applications. Therefore, we used a stereoscopic video frame obtained from a stereoscopic video on Internet as our test data. The frame is shown in Figure 1. The dimensions of the frame is 800x600 pixels; that is, left and right images contained in this stereoscopic video frame are with dimensions 400u600 pixels. After embedding an 1024-bit watermark into the image, JPEG compression, JPEG2000 compression, Gaussian filtering and sharpening operation are applied to evaluate the robustness of the proposed method. The experimental results are shown in Table 1.
As shown in Table 1, the proposed method would resist to JPEG compression, JPEG2000 compression, Gaussian filtering and sharpening. Therefore, the proposed method can be used for digital right protection. Moreover, the watermarking scheme proposed in [1] can be detected progressively, so the method described in this paper is also a progressive watermarking method. Thus, this method is suitable for digital right management on the web sites.
IV. CONCLUSION In this paper, we proposed a stereoscopic video frame watermarking scheme, which can immediately extract watermark data from a portion of a stereoscopic video frame while a stereoscopic frame is receiving. Moreover, the proposed scheme has the following additional properties: 1. Progressive detection: when partial image contents are received, watermarking detection can start to check whether or not a suspected watermark exits. 2. Early decision: watermark detector can make an early decision of whether a suspected watermark exists or not, without waiting to see the whole image contents. 3. Comparing to the watermark embedding process, watermark detection process is more efficient. The proposed scheme optimizes robustness of watermarks and quality of images in the watermark embedding process. Furthermore, the scheme simplifies the watermark detection process to reduce the computational cost. In the future, we will develop stereoscopic video watermarking methods based on the proposed method to prevent stereoscopic video from being pirated through on-line video sharing sites. REFERENCES [1] Y. H. Chen and H.-C. Huang, "A Progressive Image Watermarking Scheme for JPEG 2000," Int'l Conf. on Intelligent Information Hiding and Multimedia Signal Processing, Piraeus-Athens, Greece,2012. [2] Ingemar Cox, Matthew Miller, and Jeffrey Bloom, Digital Watermarking: principles and practice, Morgan Kaufmann, 2002. [3] Maneli Noorkami and Russell M. Mersereau, "A Framework for Robust Watermarking of H.264-Encoded Video With Controllable Detection Performance," IEEE Transactions on Information Forensics and Security, vol. 2, no. 1, March 2007. [4] Jun-Yan Hsieh and Long-Wen Chang, "A high capacity watermarking for H.264/AVG based on frequency weighting," 2006 International Symposium on Intelligent Signal Processing and Communication System, Tottori, Japan, 2006. [5] Maneli Noorkami and Russell M. Mersereau, "Digital Video Watermarking in P-Frames With Controlled Video Bit-Rate Increase," IEEE Transactions on Information Forensics and Security, vol. 3, no. 3, September 2008. Original frame Image A Sub-image b Sub-image a Image B Sub-image c Sub-image d Watermark Embedding Process
Figure 2 An example of dividing a stereoscopic video frame into sub images during watermark embedding. TABLE I EXPERIMENTAL RESULTS OF WATERMARK DETECTION AFTER JPEG COMPRESSION, JPEG2000 COMPRESSION, GAUSSIAN FILTERING AND SHARPENING. Symbol PSNR % of Correct bits JPEG Compression 30.41 66.4% JPEG2000 Compression 32.02 69.04% 3 u 3 Gaussian Filtering 28.32 70% Sharpening 24.18 98.63% 2013 IEEE 17th International Symposium on Consumer Electronics (ISCE) 296