Video Cut Detection without Thresholds

Many video cut detection algorithms have been proposed in the literature, but in most approaches several parameters and thresholds have to be set to achieve good detection results. In this paper, we present a new unsupervised learning approach to classify time series of frame disparity values into cuts and non-cuts without any thresholds and parameters. The sliding window size used in this approach is individually estimated for each video utilizing an adequate measure for clustering quality. Comprehensive experiments using the MPEG-7 test set show that our approach outperforms two recent high-quality cut detection algorithms.