A cinematic-based framework for scene boundary detection in video

Abstract

Most current video retrieval systems use shots as the basis for information organization and access. In cinematography, scene is the basic story unit that the directors use to compose and convey their ideas. This paper proposes a framework based on the concept of continuity to analyze video contents and extract scene boundaries. Starting from a set of shots, the framework successively applies the concept of visual, position, camera focal distance, motion, audio and semantic continuity to group shots that exhibit some form of continuity into scenes. The framework helps to explain the principles and the heuristics behind most cinematic rules. The idea is tested using the first three levels of continuity to extract the scenes defined using the most common cinematic rules. The method has been found to be effective.