基于人眼视觉特性与ASIFT的多尺度变换域视频水印算法研究

In recent years, with the rapid development of Internet and multimedia technologies, media products such as audio, image, video and animation have been published via the Internet. However,the side-effects of using web are obvious, anyone with Internet access can easily retrive the digital original works and then manipulate or redistribute the products without consent from the original authors. As the result, researching in digital copyright protection for multimedia data has becoming a hot research topic of the information security area and gradually drawing more attention to it. This project will be focused on the several stages. Firstly, we will research how to establish the human visal masking of the three-dimensional motion characteristic in video sequence, according to the existing video coding standard. This human visual masking could be calculated the embedding strength of the watermarking according to the content of the video sequence. And then, to improve the robustness and invisibility of the video watermarking, the video watermarking algorithms based on the Human Visual System and video motion characteristics in multiscale transform domain are proposed. Finally, because ASIFT algorithm is the improvement of the SIFT and can perfectly defend large affine transformations. We introduce ASIFT to defend the common attacks of video watermarking,especially the geometrical attacks. a video watermarking algorithms based on the Human Visual System and ASIFT in multiscale transform domain is proposed. the algorithm use the feature points and feature vectors of ASIFT to correct the video frames, which are subjected to geometrical attacks. In the algorithm, the local and global watermarking are combined. The local watermarking is used to detect the singals of watermark and the gobal watermarking is used to extract the information of watermark. It could accurately extract the watermarking signals when the video sequence is suffered the normal attacks,especially the geometrical attacks on the condition of good transparency of watermark. This research project uses a combination of the simulation and experimental methods to evaluate the proposed models and the algorithms.

本项目以基于ASIFT特征点模板的抗几何攻击视频水印算法为主要研究方向，提出人眼视觉特性与ASIFT相结合的多尺度变换域视频水印算法。本项目，首先对视频序列中三维运动特征的人眼视觉掩蔽模型做深入研究，从而获取基于视频序列的人眼视觉掩蔽特性阈值计算模型，使用该模型可根据视频序列内容，自适应计算水印的嵌入强度值。其次，为增强水印的鲁棒性和透明性，提出在多尺度变换域内结合视频运动特征与人眼视觉掩蔽特性的视频水印算法。最后，提出基于人眼视觉特性与ASIFT相结合的多尺度变换域视频水印算法，算法利用ASIFT抗仿射变换能力，对遭受几何攻击的视频帧进行校正，然后，利用局部水印与全局水印相结合，通过局部水印进行水印盲检测，全局水印进行水印信息盲提取。使算法在保证水印透明性较好的条件下对常见的视频水印攻击，特别是几何攻击具有较强的鲁棒性。项目采用仿真和实验相结合的方法对提出的的模型及算法进行评价

通过本项目，我们采用基于混合高斯建模分析算法与模糊C均值聚类分析算法，根据视频内容，对视频序列中的运动对象与非运动对象进行分割，自适应计算得到全局最优的运动阈值的判别算法；研究利用局部熵模型与多尺度变换相结合实现视频水印隐藏，通过局部熵模型自适应地计算每个方块的水印最大嵌入强度。利用嵌入算法，自适应地进行水印嵌入，以提高水印的不可见性和鲁棒性；从SIFT算法入手，利用SIFT提取的特征点具有较好的稳定性和重现性，对图像处理攻击及几何攻击均具有一定的不变性，利用这一特性，对视频水印中常见的几何攻击进行校正。然后设计一个基于SIFT的双水印系统用于抵抗常见的几何攻击算法；通过将特征点匹配的多种算法进行比较，最终采用ASIFT良好的抗几何攻击特性，提出基于仿射尺度不变特征变换(ASIFT)与奇异值分解(SVD)相结合的Contourlet域视频双水印算法；利用奇异值的旋转不变性和抗扰动性设计一个鲁棒水印，可保证视频数据的稳定与鲁棒性，增强视频数据抗几何攻击的能力。由于压缩感知可将细微的差值放大，并利用少量的数据还原原始图像，所以本项目提出基于压缩感知的脆弱水印可有效定位图像的篡改，并能对水印图像进行适当修复，将两个水印共同嵌入到视频内部，从而实现对视频数据双水印算法的嵌入；结合本项目组所隶属贵州省重点实验室这一优势，本项目组针对医学图像的信息隐藏技术进行研究，并已提出一部分算法原型，已进行仿真实验验证，针对在肺部和脑部病变诊断中，病变情况等关键信息大都出现在医学影像的软组织区域中。为有效保护肺部和脑部医学影像的软组织区域的信息完整性不受破坏，提出一种基于IWT变换域的双水印信息隐藏算法。实验结果表明，本项目中所提的多个水印算法在保证水印透明度较好的情况下，对常规几何攻击均具有较强的鲁棒性，具有一定的理论研究价值和实用价值。