Non-rigid object segmentation in video sequences using Markov random field

Abstract

The paper presents a spatio-temporal segmentation algorithm for non-rigid objects in image sequences. We use random and unpredictable characteristics of non-rigid objects in the algorithm. The algorithm consists of three steps: spatial segmentation by MRF (Markov random field) modeling, temporal segmentation by velocity vector, and the fusion of spatial and temporal segmentation. The presented algorithm has good performance in the segmentation of a non-rigid object with large deformable rate. It can be used as an effective non-rigid object segmentation algorithm for automatic VOP (video object plane) generation. We have demonstrated the efficiency of the presented method through experimental results.