基于时空标记场最大后验概率的多视频对象分割算法

Multiple Video Object Segmentation Based on Maximization of the A Posteriori Probability of Spatio-Temporal Label Field

该文提出了一种基于贝叶斯框架的时空标记场最大后验概率的多视频对象分割算法,根据视频序列帧间 (时间域)和帧内(空间域)信息的不同特点,建立基于多个对象分割标记场的最大后验概率公式,并导出其最小能量函数,通过求解最小能量使其分割标记的后验概率达到最大。最小能量的优化求解用迭代条件模式(ICM)方法, 初始分割标记场用矢量直方图法得到。实验结果表明,该文提出的算法对存在局部遮挡的多运动对象分割是有效的。

This paper presents a novel multiple object segmentation algorithm based on a Bayesian framework. According to the characteristic of the intra-frame and inter-frame （spatial and temporal） information, a representation of Maximization of the A posteriori Probability（MAP） of spatio-temporal label field is proposed. So a minimization of energy function is obtained. The optimization of solution is carried out by iterated Conditional Mode（ICM） method. The initial segmentation label fields is gotten using vector histogram. The experimental results show that the algorithm is effective to multiple object segmentation with partial occlusion.