摘要
显著度检测在计算机视觉中应用非常广泛,图像级的显著度检测研究已较为成熟,但视频显著度因其高度挑战性研究相对较少。该文借鉴图像级显著度算法的思想,提出一种通用的空时特征提取与优化模型来检测视频显著度。首先利用区域协方差矩阵构造视频的空时特征描述子,然后计算对比度得出初始显著图,最后通过联合前后帧的局部空时优化模型得到最终的显著图。在2个公开视频显著性数据集上的实验结果表明,所提算法性能优于目前的主流算法,同时具有良好的扩展性。
Visual saliency is widely applied to computer vision. Image saliency detection has been extensively studied, while there are only a few effective methods of computing saliency for videos owing to its high challenge. Inspired by image saliency methods, this paper proposes a unified spatiotemporal feature extraction and optimization framework for video saliency. First, the spatiotemporal feature descriptor is constructed via region covariance. Then, initial saliency map is computed by the local contrast of the descriptor. Finally, a local spatiotemporal optimization framework considering the previous and next frames of the current one is modeled to obtain the final saliency map. Extensive experiments on two public datasets demonstrate that the proposed algorithm not only outperforms the state-of-the-art methods, but also is of great extendibility.
出处
《电子与信息学报》
EI
CSCD
北大核心
2016年第7期1586-1593,共8页
Journal of Electronics & Information Technology
基金
国家自然科学基金青年基金(61501509)~~
关键词
视频显著度
区域协方差
局部对比度
局部空时优化
Video saliency
Region covariance
Local contrast
Local spatiotemporal optimization