摘要:视频显著性检测是计算机视觉领域的一个热点研究方向，其目的在于通过联合空间和时间信息实现视频序列中与运动相关的显著性目标的连续提取.由于视频序列中目标运动模式多样、场景复杂以及存在相机运动等，使得视频显著性检测极具挑战性.对现有的视频显著性检测方法进行梳理，介绍相关实验数据集，并通过实验比较分析现有方法的性能.首先，介绍了基于底层线索的视频显著性检测方法，主要包括5类：基于变换分析的方法、基于稀疏表示的方法、基于信息论的方法、基于视觉先验的方法和其他方法.然后，对基于学习的视频显著性检测方法进行了总结，主要包括传统学习方法和深度学习方法，并着重对后一类方法进行了介绍.随后，介绍了常用的视频显著性检测数据集，给出了4种算法性能评价指标，并在不同数据集上对最新的几种算法进行了定性和定量的比较分析.最后，对视频显著性检测的关键问题进行了总结，并对未来的发展趋势进行展望.



Abstract:As a hot topic in computer vision community, video saliency detection aims at continuously discovering the motion-related salient objects from the video sequences by considering the spatial and temporal information jointly. Due to the complex backgrounds, diverse motion patterns, and camera motions in video sequences, video saliency detection is a more challenging task than image saliency detection. This paper summarizes the existing methods of video saliency detection, introduces the relevant experimental datasets, and analyze the performance of some state-of-the-art methods on different datasets. First, an introduction of low-level cues based video saliency detection methods including transform analysis based method, sparse representation based method, information theory based method and visual prior based method, is presented. Then, the learning-based video saliency detection methods, which mainly include traditional methods and depth learning based methods, are discussed. Subsequently, the commonly used datasets for video saliency detection are presented, and four evaluation measures are introduced. Moreover, some state-of-the-art methods with qualitative and quantitative comparisons on different datasets are analyzed in experiments. Finally, the key issues of video saliency detection are summarized, and the future development trend is discussed.



PDF全文下载地址:

http://jos.org.cn/jos/article/pdf/5560