A Novel Approach of Generating Hierarchical Video Abstraction

程文刚,须德

DOI: https://doi.org/10.3969/j.issn.1006-8961.2004.01.019

2004-01-01

Abstract:ion is a short summary of the content of a longer video document, which helps to enable a quick browsing of a large collection of video data and to achieve efficient content access and representation. In this paper, we propose a novel approach of generating video abstraction at two levels, namely, the shot level and the scene level. The hierarchical video abstraction can facilitate video browsing and retrieval at different granularities. Firstly, the video stream is segmented into shots. A new key frame extraction algorithm, which does not rely on threshold, is put up to extract key frames from shots based on the content variation. An updated time-adaptive algorithm is used to group the shots into scene. Based on the defined shot similarity formula, the video scene structure is constructed after shot clustering and adjustment. Representative frames are extracted from the key frames of each scene using the method of generating Minimum Spanning Tree. Key frames and representative frames can represent the content of shot and scene, respectively. The sequence of key frames and representative frames is the two-level video abstraction. Experiments based on real-world movies show that the method above can provide users with better video summary at different levels.