Video scene segmentation based on multiview shot representation
Jeong-Woo Son, Sang‐Yun Lee, Soyoung Park, Sun-Joong Kim · 2016
A video is composed of a set of scenes containing semantics in the video. Since there have been existed user needs to consume scenes with respect to its semantics, scene segmentation has been focused by both industrial and research domains. This paper proposes a novel scene segmentation method. The proposed method is designed to keep information in multiple representations of shots as much as possible. Shots are firstly tied with adaptive multiview spectral clustering (AMSC). AMSC is designed to cluster data by preserving information among multiview. Clustered shots are then redefined as scenes through resolving overlap links in clusters.