利用图像不连续特性的溶解型镜头检测算法

Discontinuity-Driven Shot Detection Algorithm for Dissolving Shots

镜头边界检测是视频分析中的关键技术.传统的镜头边界检测算法只在切变或淡入/淡出等较为简单的镜头转变类型上有较好的检测性能,而针对溶解型镜头帧尚未有成熟的算法.为此,提出一种基于图像不连续性特性的溶解型镜头帧检测算法.首先提出一种基于图像不连续的特征,来衡量每帧图像相邻图像块之间的相似程度;然后利用支持向量机对特征数据进行预分类,并通过纠正错误分类帧和延伸溶解型镜头帧长度来修正分类结果;最终完成对溶解型镜头帧的检测.在TRECVid数据库上的实验结果表明,与传统的溶解型镜头帧检测算法相比,文中算法更能准确、有效地检测出溶解型镜头帧.

Detecting shot boundary plays an essential role in video analysis.Conventional methods work well when the shot is cut or faded in/out,while are inefficient for dissolving shots.This paper proposes a novel detection algorithm that leverages the image discontinuity features to characterize the dissolving behaviors of shots.The algorithm consists of three stages.First,the image discontinuity features are employed to measure the proximities between of neighboring image blocks in each image frame.Then,the features are classified by means of the SVM classifier.Finally,misclassified results are corrected,and dissolving shots are extended.The experiments on TRECVid dataset verified that the proposed algorithm can achieve much higher accuracy and robustness compared with traditional methods.