基于视频编辑模型的视频淡入、淡出和叠化的检测

FADE AND DISSOLVE DETECTION BASED ON VIDEO EDITING MODEL

该文基于视频编辑模型,推导出几个能用于检测镜头淡入、淡出和叠化的重要特征:方差、边界强度、有效平均梯度和双重色度差。通过比较,我们选取了方差序列和双重色度差特征作为检测依据,首先利用方差序列检测淡入、淡出和可能的叠化位置,以保证取得高的查全率,然后利用双重色度差序列确认叠化以提高检测准确率。由于检测对DC序列进行,算法计算时间复杂度比较低。

Based on video editing model, several important characteristics that can be used for detecting fade and dissolve, such as the variance, edge intensity, Effective Average Gradient (EAG) and Double Chromatic Difference (DCD) of video sequence are derived. After comparison, the variances are used to achieve robust fade detection. It is approved that in terms of recall rate for dissolve detection, the variance, edge intensity and EAG have similar efficiency. So, the variances are used for selection of possible positions of dissolve transitions, which guarantees high recall rate, and the DCD are then used for further confirmation to improve precision rate. Since the detection is carried out on DC sequence, the computation complexity of this method is relatively low.