Abstract 

A video stream consists of a number of shots each of which has the boundary property, such as cut, fade, dissolve, wipe, etc. The shot boundary such like cut is detected easily through any previous works. However, the videos with more than two types of shot and the large motion of camera or objects have difficulties in extracting the boundary between the adjacent shots. False alarms are increased in such a shot with camera and object movements. This paper proposes the shot detection algorithm which accentuates edit constancy effects while suppressing motion effects using low pass filtering in histogram space. Edit constancy effects are defined as shapes of cut or fade/dissolve in low pass filtered frame differences signal. And this paper also presents the shot representation method selecting key frames effectively based on contents within a shot.