一种新的自适应镜头边界检测算法
A New Method of Adaptive Shot Boundary Detection
-
摘要: 针对镜头边界检测中存在的检测效果和检测阈值依赖人工经验的问题,提出了一种新的自适应镜头边界检测算法. 该算法以视频帧窗口中亮度帧差到帧差均值的距离作为切变检测特征,根据该距离值的变化程度自动选取切变最优化阈值;以视频帧序列中的颜色帧差为渐变检测特征,对该帧差进行数量级量化,直接利用量化结果判定渐变,实现了镜头检测的自适应. 实验结果表明,该算法对切变的查准率和查全率分别为96.74%和97.41%,对渐变的查准率和查全率分别为92.34%和89.67%,对于不同类型的视频检测效果稳定.Abstract: Aiming at overcoming the difficulties of effect and the problem of threshold ensuring in shot boundary detection, a new method of adaptive shot boundary detection is proposed. For the shot cuts detection, distance values between brightness-differences and their average are used as detected features, and the threshold automatically chosen by the variety of the values. For the gradual shot changes detection, color-differences are used as detected feature, by quantification of color-differences according to their magnitude, gradual shot changes can be detected directly. The experimental result showed that the precision and recall are 96.74% and 97.41% for shot cuts detection; 92.34% and 89.67% for gradual shot changes detection. Besides these, the method is stable with different video types.
下载: