A New Segmentation Method for Broadcast Sports Video

Abstract

We introduce a novel method, named S-CRP (Segmentation based on distance dependent Chinese Restaurant Process), to segment broadcast sports videos into semantic shots. S-CRP employs distance dependent Chinese Restaurant Process (DCRP) using two segmentation criteria, namely appearance and time distances. It takes advantage of the customer (frame) assignments in DCRP and is able to reduce the negative effect of noisy frames without the use of domain knowledge and more sophisticated classifiers. In addition, we find that the conventional performance evaluation metrics are unable to reflect the quality of the segmentation properly. We introduced a new performance metric, namely Levenshtein distance Ratio, which gives a more accurate measure of how well the segmentation result can match the original video structure.