Research On Methods Of Video Watermarking Resisting To Geometric Attacks

With the emergence of massive digital videos, copyright protection has become a hotspot problem urgent to solve in the multimedia field. Video watermarking is one of the efficient methods to solve this problem. Most of the present video watermarking technologies can resist conventional signal processing attacks such as noise,however,it is less robust to geometric attacks. Geometric attacks can destroy the synchronization between watermark and video and influence the robustness of the watermarking greatly, therefore geometric attack is one of the fatal attack methods to video watermarking methods. How to resist geometric attacks is the focus and difficulty in current video watermarking field. To solve this problem, this paper proposes two video watermarking methods resisting to geometric attacks.First, a watermarking method based on surfacelet transform is proposed in order to resist geometric attacks in this paper. According to the space-time three dimension characteristic of video signals, the surfacelet transform is employed to perform three-layer decomposition of the video signals. In order to enhance the invisibility of watermark and in consideration of the statistical attribution of surfacelet transformation coefficients in subband direction, the original binary watermark sequence is mapped to the pseudo-random noise, which obeys zero mean Gaussian distribution. The encrypted watermark is embedded into surfacelet transformation coefficients of subband direction with high energy. When detecting watermark, SURF operator is used to extract feature points of video frame and the parameters of the geometric attacks are estimated by mapping the feature points, then Re-synchronization between watermarking and video can be gained by correcting distortion video. The relativity between the coefficients of the embedded location and watermark is employed to detect watermark and threshold value is obtained by the Neyman-Pearson criterion. Experimental results show that this watermarking method has strong robustness to the geometric attacks such as rotations, translations, scales and cropping etc.Second, a video dual watermarking method, which can effectively resist geometric attacks, is proposed in this paper. A geometric deformation invariant, namely the low-dimensional manifold of video, is mined. Geometric invariance can be theoretically derived and experimentally demonstrated. First of all, zero-watermark or dynamic watermark can be generated online according to low-dimensional manifolds of different video shots. The watermark is embedded into the IF DCT coefficients of AVS predict residuals with large energy in order to balance the robustness and transparency. The three-dimensional space-time characteristics of video is studied deeply in this paper, then video motion information is effectively quantified and characterized by low-dimensional manifold of video. And the visual dynamic masking model, including movement features, brightness features and texture features, is also established to adaptively control the watermark embedding strength. Experiments and analysis show that this method can effectively resist center cutting, irregular cutting, row cutting, rotation, scaling, translation and other geometric attacks with high intensity as well as combined attacks, such as rotation and corner cutting, center cutting and rotation etc. It also has strong robustness to the conventional video signal processing and various video attacks.Finally, the summary of this paper is given and the future direction of the research is presented.