Improved Video Object Segmentation Method Based on Background Registration
In allusion to the common occlusion problems in video segmentation methods, a novel improved video object segmentation method based on background registration is proposed. At first, the high frequency message and most noise of the images were eliminated by 1-D wavelet transform. Then, calculate the front background modeling and the back background modeling by comparing the static index and the static threshold, and employ the secondary frame difference to optimize the front background modeling and the back background modeling, then the initializing background modeling can be obtained. Accumulated frame difference was used to optimize the initializing background modeling to reduce the probability of background that is misjudged in foreground, which makes further improvement on the segmentation results. Last pick-up the video object is obtained by mathematic morphologic method. Experimental results show that the background registration method was very non-sensitive to static threshold and has strong robustness. The algorithm can overcome effectively the occlusion and quickly giving accurate segmentation results from the sequence with static background for video object segmentation, its also can preferable the method in spatial accuracy.
video segmentation background registration accumulated frame difference
Wenxiu Fu Xingmin Wang Zhongtian Man Bo Fan Bing Li
School of Electronic and Information Engineering Beijing Jiaotong University Beijing, China
国际会议
重庆
英文
866-870
2011-08-20（万方平台首次上网日期，不代表论文的发表时间）