基于形态学重建和边界融合的视频对象分割方法研究  

Research on Video Object Segmentation based on Morphological Reconstruction and Boundary Fusion

作　　者：寇万里 车嵘 严丽娜 KOU Wan-li;CHE Rong;YAN Li-na(Information and Communication Institute Experimental Training Base,National University of Defense Technology,Xi'an Shaanxi 710106,China)

机构地区：[1]国防科技大学信息通信学院试验训练基地,陕西西安710106

出　　处：《通信技术》2018年第4期825-828,共4页Communications Technology

摘　　要：通过对现有的视频序列中运动对象分割算法的研究,在时空融合的框架指导下,提出了一种基于形态学重建和边界融合的视频对象分割方法。具体地,空间域利用形态学重建及形态学梯度阈值判别的改进分水岭算法,有效抑制了"过分割"现象;时间域采用变化检测的方法来初步确定运动区域,采用高阶统计量的方法进行高斯检验,有效去除了视频序列存在的背景噪声;提出了基于边界的四阶矩,用以滤除噪声并进行时空融合,较最初的四阶矩方法,大幅提升了运算效率。Through the research of the moving object segmentation algorithm in the existing video sequences,a video object segmentation method based on morphological reconstruction and boundary fusion is proposed under the guidance of the spatiotemporal fusion framework.Specifically,an improved watershed algorithm that uses morphological reconstruction and morphological gradient thresholds in the spatial domain effectively suppresses the“over-segmentation”phenomenon.In the time domain,the method of change detection is used to determine the motion area initially,and the Gaussian test is performed using high-order statistics.This effectively removes the background noise in the video sequence.A fourth-order moment based on the boundary is proposed to filter noise and spatio-temporal fusion.Compared with the original fourth-order moment method,the operation efficiency is greatly improved.

关 键 词：视频对象分割 形态学重建 时空融合 高斯检验 

分 类 号：TN911.73[电子电信—通信与信息系统]