摘要
镜头边界检测是视频检索的首要问题,镜头转换分为突变和渐变,镜头边界的检测结果直接影响视频检索的准确度,针对这个问题,提出了在压缩域视频中进行镜头边界检测常用的两类方法:一类是基于I帧DC系数的方法;另一类是基于聚类的方法。前者先利用I帧的DC图进行镜头的粗略分割,再分别运用基色调、宏块信息和运动矢量进行精确分割;后者聚类法克服了帧的无序性。实验结果表明,第一类压缩域镜头边界检测的方法之于镜头的渐变检测效果普遍不理想,但是计算较第二类算法简便,第二类方法对渐变镜头的检测效果好于第一类,有效克服无序性是一种改进。
Shot boundary detection is the first important problem towards video inspection. Lens transfbrmation is classified into abrupt-change and gradual-change, this paper summarizes the commonly used two kinds of methods in compressed domain of video shot boundary detection：one kind is based on I frame DC coefficient method; another kind is based on the method of clustering. The former makes rough segmentation by using DC map of I frame diagram of the lens, then makes precise segmentation by using base color, macro block information and motion vector respectively. The latter overcomes disorder of frame. The research results show that the first kind of method of boundary detection in corn- pressed domain lens gradient testing general effect is not ideal, but the calculation is simpler than the second algorithm, the second category of methods for detecting the gradient lens effect is better than the first category, effectively overcome the disorder is a kind of improvement.
出处
《自动化技术与应用》
2015年第4期84-89,共6页
Techniques of Automation and Applications
关键词
镜头边界检测
基色调
宏块
运动矢量
聚类
shot boundary detection
base colors
macro block
motion vector
clustering
