【作者】 周志立；

【导师】 杨高波；

【作者基本信息】 湖南大学 ， 通信与信息系统， 2010， 硕士

【摘要】 随着数字视频技术的发展,数字视频数据量呈爆炸性增长,因此对数字视频的检索、管理和版权保护的产生了迫切的需求。基于内容的视频拷贝检测(简称视频拷贝检测)技术在视频信息管理、过滤和版权保护等方面有着重要意义。采用视频拷贝检测技术可以快速有效地找出数字视频的拷贝进行版权保护,对视频信息检索得到的结果进行过滤和排序,或者监测一段数字视频的版权和播出状况等。近年来,视频拷贝检测成为一个新兴的前沿研究领域。本文分析现有视频拷贝检测算法存在的问题,在像素域和压缩域两个方面对视频拷贝检测技术展开研究,提出了两种具有较好检测性能的算法。论文的主要工作如下：第一,介绍视频拷贝检测的系统结构和拷贝视频的种类；对现有的视频拷贝检测算法的研究现状、研究方向和研究重点进行了阐述；介绍视频镜头检测、关键帧提取、视频特征提取和特征匹配等视频拷贝检测的关键技术。第二,为了解决分块数和对添加边框攻击鲁棒的矛盾,本文提出一种二次分块的时空联合方法。本方法首先将查询视频和测试视频序列的每帧图像分别分成2×4块和4×2分块两种模式,在两种模式下分别提取排序特征向量并计算特征相似度,取较大相似度作为计算空域特征相似度。接着,联合时域特征相似度得出视频相似度值。最后,根据视频相似度进行视频匹配。由实验结果可知,该算法不仅保证针对添加边框和多种拷贝行为鲁棒性好并具有较高检测精度。第三,提出一种结合视觉感知的压缩域视频拷贝检测算法。为了达到快速检测的目的,通过直接对DCT域提取视频特征并建立索引进行二级匹配；在特征提取时充分考虑人眼视觉感知的特性,根据人眼视觉对空域区域和频域的敏感度不同,分别提取Ⅰ帧的结合视觉感知的DC系数和AC系数排序特征,从而保证算法抵抗拷贝攻击能力更强。实验结果表明,与现有的相关算法相比,该算法不仅检测精度高,针对各种拷贝攻击鲁棒性强,并显著提升了检索速度。更多还原

【Abstract】 With the development of digital video technology, the data of digital video increases exponentially, therefore, there is an urgent need of the indexing, managing and protecting intellectual property rights (IPR) of digtal video. The technology of content-based video copy detection (called video copy detection for short) plays an important role in the aspects of video retrieval, video filter and protecting IPR. By using the technology of video copy detection, we can effectively find out the copy of digital video to protect intellectual property rights, percolate and sort the results of video retrieval, or supervise the copyrights of a digital video and its broadcasting conditions. Recently, video copy detection technology becomes a focus research.Analyzing the problems of present video copy detection algorithm, this essay makes researches on video copy detection technology from the two aspects of pixel domain and compressed domain, and proposes two effective methods. The major work of this essay are:(1) Firstly, the paper introduces video copy detection technology and some kinds of copy video; Secondly, describes the present, direction, key points of the technique. Third, describes the technical of shot detection key frame and the feature of key frame extracting and video matching.(2) In order to solve the contradiction between the number of subblocks and the robust of letter-box and pillar-box formats conversions, this essay employs the method of double deblocking based Spatiotemporal video copy detection. Fistly, the frames of video sequence are divided into two models of 2×4 and 4×2; Under these two models, the method of spatiotemporal is employed to calculate the similarities of spatio features between original video and test video and uses the higher similarity as spatio feature similarity. And then, it mesures the video similarity by combining spatio feature similarity with temporal feature similarity. The results of the experiment show that this method not only ensures the robust of letter-box, pillar-box formats and other formats conversions, and also has high precision.(3) It proposes a method of compressed domain video copy detection based upon visual Perception. In order to improve the speed of the detection, it directly extracts feature from dct domain, and uses the two-level hierarchical detection scheme with creating index to reduce the time of process. In the process of extracting feature, by the different human sensitivity of different areas and frequency, extracts the ordinal feature of DC and AC coefficients. The experiment results show, compared with the previous algorithm, the algorithm can enhances the robustness of multiple copy video attack and improves the detection speed obviously with the higher detection precision.更多还原