COLOR VIDEO SEGMENTATION BY TEMPORAL LINKING OF IRREGULAR PYRAMIDS

摘要

This paper presents a new spatiotemporal segmentation and motion estimation technique for color video sequences. It relies on building irregular pyramids over consecutive frames. The pyramids are built and pruned in a non-iterative bottom-up way to split the image into homogeneous regions. After that, a hierarchical merging stage is performed and the resulting regions are easy to match between consecutive pyramids. Also, to overcome the problem derived from the presence of fast moving objects, a prediction motion model is introduced. The main advantages of the proposed algorithm are that it is computationally fast and it preserves region connectivity.