融合多尺度信息和MRF的序列图像运动目标分割  

Sequences Image Moving Object Segmentation Fusion of Multiscale Information and MRF

作　　者：师冬霞 夏平[1,2] 雷帮军[1,2] 任强[1,2] Shi Dongxia;Xia Ping;Lei Bangjun;Ren Qiang(Hubei Key Laboratory of Intelligent Vision based Monitoring for Hydroelectric Engineering,Three Gorges University,YiChang 443002,China;College of Computer and Information Technology,Three Gorges University,YiChang 443002,China)

机构地区：[1]三峡大学水电工程智能视觉监测湖北省重点实验室 [2]三峡大学计算机与信息学院,湖北宜昌443002

出　　处：《信息通信》2019年第8期15-18,共4页Information & Communications

基　　金：国家自然科学基金（联合基金）重点项目（U1401252）;省重点实验室开放基金项目（2018SDSJ07）

摘　　要：提出融合小波域多尺度信息和MRF模型的序列图像运动目标分割算法。对视频每帧图像进行小波多分辨率分析,融合每一尺度的特征信息,在此基础上构建每一尺度MRF模型的特征场和标记场;标记场采用Potts模型建模,同一尺度观测特征场采用混合高斯模型建模,同标记的特征场采用高斯模型进行建模,相邻尺度间标记场用一阶Markov转移概率描述;最后,利用迭代条件模式(ICM)实现MRF模型中后验分布函数最优,完成运动目标分割。实验结果表明,该算法能较好的提取运动目标信息,在固定场景的视频监控中具有一定的适用价值。Sequence image moving object segmentation algorithmwhich based on multiscale information fusion and MRF model is proposed.Wavelet multi-resolution analysis on each frame of the video is performed,and the characteristic information of each resolution scale is extracted.On that basis,the characteristic field andmarker field of eachMRF model are constructed;marker field ismodeled by using the Potts model,the same scale observation characteristic field is modeled by using Mixed Gauss model,Characteristic field of the same mark is modeled by using Gauss model,adjacent scale marker field is described by using first orderMarkov transfer probability;Finally,the posterior distribution function optimumof theMRF model is implemented by using the iterative model(ICM),and the moving target segmentation is completed.The experimental results showthat the proposed algorithmcan extract the moving object information better and have a certain application value in the video surveillance of the fixed scene.

关 键 词：运动目标分割 MRF模型 多尺度信息融合 标记场 迭代条件模式 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]