一种时域和梯度域相结合的视频对象提取算法  

A New Algorithm for Video Object Extraction Based on Temporal Domain and Gradient Domain

作　　者：封春升[1] 郝爱民[1] 何兵[1] 王莉莉[1] 

机构地区：[1]北京航空航天大学虚拟现实新技术教育部重点实验室,北京100083

出　　处：《中国图象图形学报》2008年第3期494-498,共5页Journal of Image and Graphics

基　　金：国家高技术研究发展计划863项目(2004AA115130);国家重点基础研究发展计划973项目(2002CB312105);国家自然科学基金重点项目(60533070)

摘　　要：为了从复杂背景中精确地提取出视频对象,提出了一种融合时域和梯度域信息的视频对象提取算法,可以有效地提取出复杂背景下的视频运动对象,并解决前景与背景一致情况下,背景剔除方法所带来的空洞问题。首先在时域空间中分别采用背景剔除和帧间差分方法生成初步的视频对象,并利用形态学中的二值腐蚀和膨胀方法对视频对象进行处理;然后,在梯度域空间中用Sobel算子进行视频对象边缘检测,并结合时域空间中的视频对象,生成精确的视频对象轮廓边缘;最后,采用启发式搜索方法连接视频对象轮廓边缘点,进而提取出视频对象。实验结果表明,该方法能够比较完整精确地从复杂背景中提取出视频对象。A novel approach for video object extraction fusing temporal and gradient information is proposed. Experiments demonstrate that our method can extract the video object effectively from complex background, and can resolve the vacancy phenomenon when the foreground and background are of the same color using background eliminate method. Firstly, in the temporal domain we introduced background eliminate and adjacent frame difference method to generate elementary video object, and then we used erosion and dilation morphologic method to the video object. Secondly, in the gradient domain we detected video object＇ s edge using Sobel operator. Associating with the video object generated in the temporal domain, we obtained the exact video object contour＇ s edge. Finally, we applied heuristic search method to link video object contour＇ s edge points, sequentially extracted video object.

关 键 词：视频对象提取 背景剔除 边缘检测 边缘连接 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]