Abstract: Aimed at the problem that the video segmentation method based on conditional random fields needs set empirical values to obtain optimal segmentation results,the video segmentation model was proposed to compute adaptively video neighboring relationship feature function by video features and construct adaptive parameters conditional random fields. The core idea of algorithm is to calculate different kinds of model feature function by pixel neighboring relationship in video, these feature energy functions arc constrained by conditional random field model, which is solved via Gibbs sampling algorithm to obtain globally optimal segmentation results. The experiment shows that the results of adaptive parameter algorithm are the same as one of optimal empirical parameters.

Key words: Video segmentation,Background modeling, Adaptive parameters, Conditional random fields