利用相似性比较进行视频镜头检测的方法  

Similarity Based Algorithm to Detect Video Shot

作　　者：叶军[1] 汪娥[2] 李建良[3] 

机构地区：[1]南京邮电大学应用数理学院,江苏南京210003 [2]南京农业大学经济管理学院,江苏南京210095 [3]南京理工大学理学院,江苏南京210094

出　　处：《河北理工学院学报》2007年第2期75-80,共6页Journal of Hebei Institute of Technology

基　　金：江苏省自然科学基金BK2004137

摘　　要：随着多媒体技术的发展,大量的视频数据涌现。但由于视频数据是非结构性的数据,故而,为了便于视频的检索和浏览,必须对视频进行分段,将镜头作为分段的基本单元是比较合适的。通过改变两帧图像的分块相似度的定义对镜头进行突变检测,然后利用小波变换对分块图像进行多尺度分解,以提取图像的边缘特征;最后结合这两种特征,对镜头进行渐变检测,数值试验表明,所利用的方法对各种类型的视频能够很好的检测出其中的突变型镜头和渐变型镜头。With the development of multimedia technology, the amount of multimedia video data is enlarged at an explosive speed. As videodata isn＇ t structural, we have to detach it in order to index andbrowse video. It is a good method to use shots as the basic units. We improve the similarity between two frames using the blocked gray histogram difference as features to detect the break shots. Then a multi - resolution decomposition using Hart wavelet is applied, several statistical features of the blocked frame edge based on wavelet coefficients are obtained and similarity comparison on the features is undertaken. These two kinds of similarity are combined to synthetically predict the similarity between two frames and be used to detect the gradual transitions of shots. From the results of the experiments, it is shown that our method can examin all kinds of the shot edges greatly.

关 键 词：镜头边界检测 小波变换 分块直方图 图像相似 

分 类 号：TP334.1[自动化与计算机技术—计算机系统结构]