融合颜色与运动信息的视频显著性滤波器  

Color and motion information fusion based saliency filter for video

作　　者：罗雷[1,2] 蒋荣欣[1,2] 田翔[1,2] 陈耀武[1,2] 

机构地区：[1]浙江大学数字技术及仪器研究所,浙江杭州310027 [2]浙江大学浙江省网络多媒体重点实验室,浙江杭州310027

出　　处：《华中科技大学学报（自然科学版）》2014年第2期81-85,共5页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(40927701);浙江省重点科技创新团队资助项目(2011R09021-06)

摘　　要：根据人眼视觉的特点,提出一种融合颜色及运动信息的视频显著性滤波器.该滤波器分三个步骤检测视频图像的显著性区域:首先利用超像素分割算法将视频图像划分为若干视觉近似一致的紧凑子块;其次以统一的多维高斯滤波形式计算四个测度,分别表征每个子块的颜色独立性、颜色空间分布、运动独立性以及运动空间分布;最后同样通过多维高斯滤波对四个测度进行融合以生成视觉显著性图,进而检测出视频中的显著性区域.实验结果表明:与其他现有的显著性检测方法相比,该滤波器能更准确地检测出视频中的显著性区域.Based on the characteristics of human vision system ,a color and motion information based video saliency filter was proposed .The proposed filter detected the saliency region in video sequences with three basic steps .First ,each video frame was decomposed into several compact ,perceptually u-niform elements .And then ,four measures were formulated with uniform N-D Gaussian filter and computed to describe the color uniqueness ,color spatial distribution ,motion uniqueness ,and motion spatial distribution of each element .Finally ,the four measures were also fused with N-D Gaussian fil-ter to produce visual saliency map w hich indicated the salient regions in the videos .T he experimental results demonstrate that the proposed method outperforms other state-of-the-art algorithms .

关 键 词：图像处理 视频图像 显著性 超像素 运动矢量场 高斯滤波 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]