Semantic Audiovisual Features in Video Scene Detection

Abstract

A new semantic based video scene detection technology is proposed to improve the effectiveness of video retrieval. In this study, we proposed a framework for constructing semantic audiovisual features for detecting human action. Two major components, including multimodality features and matching process, are discussed and their corresponding algorithms are presented. A pilot evaluation was conducted to compare the effectiveness of multimodality that supports semantic audiovisualbased detection. Some experimental examples are given to demonstrate the effectiveness and efficiency of proposed algorithms.