Redundant Segment Detection and Removal by Clustering of CNN Features in Sequential Video

Abstract

Video taken by ordinary users are used for a variety of purposes. In particular, video that record the photographer's daily life are used for looking back on personal activities. Since it takes the same amount of time to view a video as it was shot, there are video summarization methods that greatly reduce the viewing time by extracting only the important parts of the video. However, most video summarization methods are limited in video types and there are few effective studies for a long video without cut. In this study, we focus on redundant scenes, by detecting them based on the clustering of CNN features, and propose a method to summarize the video.