鲁棒的镜头边界检测与基于运动信息的视频摘要生成

根据基于内容的视频索引与检索等应用的需求,提出一种视频摘要生成方法.首先进行鲁棒的镜头边界检测,基于颜色直方图计算相邻帧间距离来进行初步检测,并通过分析帧间运动向量去除由相机运动引起的误检测;然后根据镜头的运动指示图将镜头分为静态镜头、包含对象运动的镜头和包含显著相机运动的镜头;最后提出镜头间基于多实例表示的距离度量方法以及聚类算法的初始化方法,采用核K-均值算法对每类镜头进行聚类,抽取每类中最靠近类簇中心的镜头作为关键镜头,将关键镜头按时间序组合起来形成视频摘要.与已有方法相比,文中方法能进行更鲁棒的镜头边界检测,识别镜头中的运动信息,并对镜头分类后进行分别处理,从而增强视频摘要的信息概括能力.

Robust Shot Boundary Detection and Video Summarization Based on Motion Information

We propose an automatic approach to reconstruct realistic 3D facial expressions from un-calibrated monocular video sequences with little constraint.By the approach,the affine rectified optical flow method is first used to track facial features automatically aligned with ASM algorithm in the first frame.Secondly,3D personalized face model and 3D expressions are reconstructed by shape from motion algorithm with the help of a generic face.At last,dynamic texture mapping is applied instead of traditional static...