论文研究 Automatic Shot Boundary Detection on Short Internet Test Suit.pdf

基于互联网短视频的镜头分类，黄一，张洪刚，视频边界检测室视频内容分析的第一步，已经有很多检测算法出现，并且能取得不错的效果。这些算法主要分为两大类：突变检测和渐变山国武技论文在丝山国武花论文在丝Thc abrupt boundary dctection could causc a grcat changc in histogram, which is casy to bcdetected. But flashlight can also cause this effect. So by only using the histogram difference todetect cut boundary will have a lot of false detection. We add a flashlight detection process todistinguish flashlight from a real cut transitionCalculating the histogram of each frame we conclude that the intensity changes from on levelto another when a cut transition occurs, then it will kccp stablc in the next shot. If flashlightappears,Anothcr kcy issuc is how to sct thc appropriate threshold to dctcrminc segment. Experimentshave shown that the threshold must be based on the variation of shots So the adaptive threshold isbeller than the global threshold. Here we use a sliding window to compute the cut threshold,which is shown as follows4)Where is the length of the sliding window is the index of first frame in current window.IS an empirical value, here we set it to 4. at the same time we calculate the variance of histogramdifference. When the variance it lager than a threshold, we say the current window is not suitableto be a data source to compute threshold山国武技论文在丝Black fr>=TsimMonotone increase/decrease->Similarit<=Tourfade山国武技论文在丝+-8CandidaterrMonotone change rate- Difference from ideal dissolve framedissDuration>TourSimilarity be tween fs and feTsImDissolvenot dissolve∑山国武技论文在丝Cuvideoprecision recall precision recall0.9250.9340.5520.6310.9510.9120.430.6810.9010.9410.50.5780.8920.9500.390.692山国武技论文在丝基于互联网短视频的镜头分类黃一,张洪刚,郭杰(信息与通信工程学院,北京邮电大学摘要:视疑边界检测室视频内容分析的笫一步,已经冇很多检测箕法岀现,并且能恥得不错的效果。这些算法主要分为两大类:突变检测和渐变检测。这篇文章结合物体的多种特征来检测视频边界。首先采用直方图之差去除明显不是镜头变化的帧,有效的减少了计算量。另外本文采用了溶解和淡入淡出的一些典型特征来检测这两类渐变镜头。本文还增加了闪光闪光检查模块以减少误检测。本文提供了一种简单有效的方式来实现镜头检测,在互联网短祝频的检测中取得了高达的准确率。关键词:模式识别与智能系统;镜头边果检测;淡入淡岀;溶解中图分类号