Movie Event Detection by Using AudioVisual Information

Abstract

This work focuses on the detection of events including 2-speaker dialog scene, ni ultiple-speaker dialog went! and hybrid scene from the daily movies, Specifically, we employ the visual information to detect all possible shot sinks by using a new method called a window-based sweep algorithm, then all detected shot sinks we categorized into 3 classes using the well-known K-means algorithm, which forms the basis of our proposed event-detection scheme. furthermore, the accompanying audio cue will also be utilized for achieving more meaningful results. Our preliminary experimental results show that by integrating audiovisual information, meaningful events could be successfully detected and extracted.