Rushes Video Summarization by Audio-Filtering Visual Features

Abstract

In this paper, we propose a video summarization system for analyzing basketball game videos.In contrast to previous video analysis technologies employing only visual and motion features to do video filtering, we add audio features to do video summarization in the system.First, we extract replay highlights by special effect detection.Next, we filter landscape shots using color range pixel and fast motion activity.Then, the corresponding audio features extracted from these landscape shots are used to identify landscape shot highlights by an SVM.Finally, we integrate the replay highlights and landscape shot highlights to complete the video summarization.From the experimental results, we find that the accuracy on the special effect detection, landscape shot extraction, and landscape shot highlight detection is very high.Thus, the final video summarization has high recall values on highlight extraction.