摘要
本文针对未进行后期编辑的视频序列设计了一个视频摘要系统。以视频帧图像的累积直方图来提取图像特征,有效地解决了光照的改变使镜头内直方图差异较大以及直方图对物体/摄像机的运动的鲁棒性有限两大问题,并结合了滑动窗口算法,有效地实现对视频序列的镜头边界检测。另外,本文还提出了一种基于镜头和图像内容的关键帧提取算法,降低计算的复杂度,并且提取的关键帧能更全面准确地体现原视频的内容。
This paper focuses on designing a system that is capable of abstracting useful video frames for archiving. Cumulative histogram is adopted to detect the edges of video frames due to its lower sensitivity to the motion of objects/camera and illumination variations than statistics histogram. Dynamic threshold based sliding-window is used to detect the shot boundaries. To get the key frames in favor of its represent- ativeness, we use the key frame extraction method based on the shots and image content.
出处
《中国传媒大学学报（自然科学版）》
2013年第1期44-51,30,共9页
Journal of Communication University of China：Science and Technology
关键词
视频摘要
直方图
镜头边界检测
关键帧
帧差值
video abstraction, histogram, shot boundary detection, key frame, frame differencing
