【作者】 向志敏；

【导师】 景晓军；

【作者基本信息】 北京邮电大学 ， 通信与信息系统， 2009， 硕士

【摘要】 电子设备的普及,使得在互联网上每天有以数以百计的视频被创造出来。同时,盗版视频也在成倍的出现,这极大地侵害了版权所有者的利益了,并对视频用户的搜索结果造成影响。在这样的背景下,视频拷贝检测技术成为热点。基于内容的视频拷贝检测是水印技术的有效补充,其基本思想是认为视频自身包含了足够多的能够用于检测拷贝的信息。通过测量从查询视频和测试视频中提取的能唯一标识视频内容的特征序列之间的距离来判断测试视频是否是查询视频的拷贝。它主要包含了镜头分割、关键帧检测、视频特征提取、特征值距离测量四部分内容。本论文对镜头分割技术和关键帧特征提取技术进行了深入研究,对基于主颜色直方图的图像特征提取方法进行了改进,提出了一种新的HSV空间的分块主颜色直方图特征值的提取方法。在镜头边界检测中,使用新提出的方法作为镜头分割的主特征值,并结合帧图像主颜色像素比率改善算法结果;视频关键帧序列相似度检测采用和镜头分割相同的主特征值,而使用基于Canny算子的边缘特征值作为辅助特征,以提高准确性。最后,通过实验表明,使用本文提出的特征值提取方法能很好检测出同一内容的视频在不同分辨率和不同格式情况下的拷贝。更多还原

【Abstract】 With the popular of electronic equipment, there are hundreds of videos and video copies are created in one day. These copies affect the video search result and against the interest of copyright holder. Video Copy Detection aims to find the copies.The most popular technology in Video Copy Detection is Watermarking, and Content-based Video Copy Detection is a complementary approach to watermarking for detecting copies of media. The primary thesis of Content-based Copy Detection(CBCD) is "the media itself is the watermark", the media contains enough unique information that can be used for detecting copies. CBCD schemes measure the distance between signatures extracted from the original media and a possible copy, which contains shot boundary detection, key frames detection, video signatures extraction and distance measure.A novel content-based video copy detection method was proposed. At first, this paper improves the domain color histogram in HSV space, and detects the shot boundary with the number of domain color pixels. Then use the same way to extract the character of sequence of key frames to detect the video copies, with the image boundary signature base on Canny detector. The result show the method can detect the video copies in different resolution and different format.更多还原