节点文献
基于push-relabel的视频对象提取算法的研究与实现
Research and Implementation of Video Object Extraction Based on Push-relabel Algorithm
【Author】 Li Yan~1 Guan Yudong~1 Wang Hongmin~2 Wang Ying~1 (1 Information and Communication Engineering Institute,Harbin Institute of Technology,Harbin 150001 2 Harbin University of Science and Technology,Harbin 150080)
【机构】 哈尔滨工业大学信息与通信工程系; 哈尔滨理工大学;
【摘要】 本文介绍了基于push-relabel算法的视频对象提取方法,采用混合高斯模型估计背景参数,将连续彩色视频图像序列映射到相应的低维的图空间,运用push-relabel算法对构造出的图进行搜索,找到视频图像的最小割。本文算法对低分辨率视频图像分割速度较快并且只需调整部分参数就可以实现对不同视频文件的分割,具有较好的通用性。实验结果表明,分割出的视频对象基本满足要求。
【Abstract】 This paper presents a video object extraction method based upon the push-relabel algorithm. Background parameters are estimated with Gaussian mixture model.In order to avoid the complex processing of continuous color video image sequences,images are mapped to corresponding graphs in lower dimension and use the push-relabel algorithm to search the minimum cut of graph.This algorithm separates low-resolution video images fast and just needs to change some parameters for different video files,which enhances the generality of the video segmentation.The simulation experiment results basicly meet requirements in practice.
【Key words】 video extraction; Gaussian mixture model; push-relabel; minimumcut;
- 【会议录名称】 2008中国仪器仪表与测控技术进展大会论文集(Ⅱ)
- 【会议名称】2008中国仪器仪表与测控技术进展大会
- 【会议时间】2008-06
- 【会议地点】中国湖南湘潭
- 【分类号】TP391.41
- 【主办单位】中国仪器仪表学会、《仪器仪表学报》杂志社、《国外电子测量技术》杂志社、《电子测量技术》杂志社