摘要
提出了一种基于时空联合分析框架的视频对象分割算法,通过改进的分水岭变换对视频图像进行帧内空间区域划分,并根据帧间运动信息和区域的空间特性得到初步的分割掩模;然后建立基于区域的马尔可夫随机场分布模型,并定义对应的Gibbs势能函数,通过迭代条件模式(ICM)方法求解得到最小化能量,从而获得稳定的分割标记场,准确地提取视频对象。实验结果表明,提出的分割算法性能优于欧洲COST211研究组所得到的分割结果。
A novel video motion object segmentation algorithm was proposed under spatial-temporal framework. First, the inter-frame image was partitioned into some regions by the improved watershed algorithm, and the initial segmentation mask was obtained after combining the spatial and temporal segmentation results. Then a model of Markov random field was built based on regions, and the corresponding Gibbs potential energy function was defined. The optimization of solution was carried out by iterated conditional mode (ICM) method. The experiment results show that the algorithm is more effective than COST 211 analysis model.
出处
《通信学报》
EI
CSCD
北大核心
2005年第6期57-61,共5页
Journal on Communications
基金
国家自然科学基金资助项目(60172020)
