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

项目名称： 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

项目编号： No.61462013

项目类型： 地区科学基金项目

立项/批准年度： 2015

项目学科： 自动化技术、计算机技术

项目作者： 李智

作者单位： 贵州大学

项目金额： 45万元

中文摘要： 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

中文关键词： 视频水印；人类视觉系统；视频运动特性；多尺度分析；ASIFT

英文摘要： 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.

英文关键词： Digital Video Watermarking;Human Visual System;Motion Characteristics of Video;Multiscale Analysis;ASIFT