A region-based MRF model for unsupervised segmentation of moving objects in image sequences

This paper addresses the problem of segmentation of moving objects in image sequences, which is of key importance in content-based applications. We transform the problem into a graph labeling problem over a region adjacency graph (RAG), by introducing a Markov random field (MRF) model based on spatio-temporal information. The initial partition is obtained by a fast, color-based watershed segmentation. The motion of each region is estimated and validated in a hierarchical framework. A dynamic memory, based on object tracking, is incorporated into the segmentation process to maintain temporal coherence. The performance of the algorithm is evaluated on several real-world image sequences.