Shot-HR: a video shot representation method based on visual features

摘要

One of the main approaches for video scene segmentation is based on shot clustering. However, shot characterization has been left in background by researches related to video segmentation. Therefore, in this work, we propose a shot representation method based on visual features. This method uses SIFT descriptor, considers all shot frames and consists in three main steps: features reduction at frame level, features reduction at shot level and match. We evaluated our shot representation method in the scene segmentation context, with videos from movies domain and using the technique proposed in this paper. We developed a comparative study with three state of art approaches based on keyframes and the results show that our method overcomes those approaches.