多媒体内容与检索技术分析报告.ppt

* * * * */152 * transformations often used for generating the video copies 左图为简单的拷贝检测，视频长度不变，仅对内容做攻击； 右图为复杂的拷贝检测，将若干不同来源的视频片段进行组合， 要求准确的判断和定位。 * * 2007年CIVR国际年会举办了第一次视频拷贝检测国际评测；2008年美国国家标准局举办的TRECVID国际多媒体内容分析评测比赛加入了拷贝检测项目，正式定义这个学术问题；表明学术界和工业界对此研究的关注达到了新的高度 SIFT: 使用了2中 spatial info。 一种是相邻区域的能量分布signature，一个是匹配后的query和candidate frame 的 SIFT 分布。 In order to reduce the illegibility of local feature, we present a method to describe its spatial information, as in Fig 4. First, the neighborhood of each local point is divided into 4 blocks (scale is 2 times of the local patch), and these blocks are sorted using their average gray level, something resembling [4]. The rank of each block is used as the spatial signature of each local point, which is compact and distinguishing. With the spatial constrain, we can effectively reduce most false matching of local features. Second, given a query keyframe with local features {q1, q2…, qn}, we will obtain several candidate matching keyframes KFi denoted by {Ci1, Ci2…, Cin}. If we divide the bounding box containing these local features into N*N blocks, then the spatial distribution histogram can be easily calculated as Figure 4. With this information, we can further reduce the false matching of local features. * * * Adaptive Image Signature. It is quite intuitive that the same set of visual features may not work equally well to characterize, say, computer graphics and photographs. To address this issue, learning methods have been used to tune signatures either based on images alone or by learning on-the-fly from user feedback. In Figure 6, we categorize image signatures according to their adaptivity into static, image-wise adaptive, and user-wise adaptive. Static signatures are generated in a uniform manner for all the images. Image-wise adaptive signatures vary according to the classification of images. The term semantic-sensitive, coined in Wang et al. [2001], reflects such a mechanism to adjust signatures, and is a major trait o