An Unsupervised Playfield Segmentation for Various Sport Videos

In this paper, we propose an unsupervised method of playfield segmentation for various sport videos. The method first applies local maximum clustering to gather feature samples into several clusters in the Cb-Cr plane. Next, a novel idea is developed to merge clusters into four color classes - Red, Green, Blue and Grey. Finally, a simple scheme of region fusion eliminates small and unimportant areas. The experimental results indicate that the method effectively segments the playfield regions in various scenes of different sport videos.