High-level soccer indexing on low-level feature space

In this paper, we propose an efficient scene clustering algorithm for soccer videos. Such scene clustering in sports videos is one of the semantic indexing techniques that can he applied to editing, summarizing, and content structuring. Our proposed method exploits a small set of audio and visual low level features that can be easily extrdcted from an MPEG compressed video, and classifies a soccer video into five predefined scene classes, from lower importance to higher importance. Based on very simple determination criteria, simulation results have shown that our method successfully performs semantic indexing for soccer videos at low computational cost.