基于频繁镜头集合的视频场景分割方法

A VIDEO SCENE SEGMENTATION METHOD BASED ON FREQUENT SHOT SET

提出一种基于全局场景特征在视频序列中寻找频繁镜头集合,并通过局部语义特征精确定位视频场景边界的视频场景分割方法。首先对分析视频进行高精度镜头分割,选取具有代表性的镜头关键帧。然后提取各镜头关键帧的全局场景特征和局部特征,并利用局部特征聚类得到的视觉词对各个镜头关键帧进行语义标注。接下来计算基于全局场景特征的镜头间相关性,结合视频场景的概念和特性,在镜头关键帧序列中寻找局部频繁出现的相关性高的镜头集合,粗略定位视频场景位置。最后利用镜头关键帧的语义标注特征精确定位视频场景边界。实验证明该方法能够准确、有效地检测并定位到大部分视频场景。

The paper proposes a video scene segmentation method that searches for frequent shot sets in video sequences on the basis of global scene characteristics as well as precisely locates video scene borders by local semantic properties.At first the analyzing video is shot split by high resolution to choose representative shot key frames.Then the global scene features and local features of every shot key frame are extracted.Then with the visual vocabulary created by local feature clustering,every shot key frame is semantically labeled.Next the relativity among shots based on global scene features is calculated.Combining the video scene concept and features,shot sets with high relativity in local frequent appearance are sought for among shot key frame sequences in order to roughly locate the video scene.At last the shot key frame semantic labeling feature is used to precisely define the video scene border.Experiments prove the method can accurately and effectively detect and locate most video scenes.