视频对象分割中基于Gibbs随机场模型的空分割结合方法

本文提出了一种基于Gibbs随机场模型的时空分割结合方法,用于视频对象的分割.该方法为每一帧图像的分割模板建立Gibbs随机场模型,将时间域分割结果作为初始标记场,空间域的分割结果作为一个图像观察场,然后利用Gibbs模型的约束条件将二者结合起来,得到该帧最后的分割标记场.实验结果表明,这种时空结合方法可以较好地避免以往的比重法过分依赖于空间域分割精度的问题.

Integration of Spatial and Temporal Segmentation Based on Gibbs Random Field Model in Video Object Segmentation

An algorithm based on Gibbs random field model for the integration of spatial and temporal segmentation results is proposed in this paper for video object segmentation.The segmentation mask of each frame is modeled as a Gibbs random field.The field is initialized as the temporal segmentation result and the spatial segmentation is provided as an observed field of the image.Finally,the mask is gotten through integrating spatial and temporal segmentation results according to the restriction of Gibbs random field model.Experimental results show that the novel algorithm can avoid the overfull dependence on the accuracy of the spatial segmentation as the usual Ratio Method does.