A Hierarchical Organization of Home Video
Y.J. ZHANG
DOI: https://doi.org/10.4018/978-1-4666-5888-2.ch210
2015-01-01
Abstract:With the progress of electronic equipments and computer technology for taking motion pictures and processing huge data, an increasing number of people now own and use camcorders to make home videos that capture their experiences and document their lives. Home video has no time limits and no restriction in content (Lienhart, 2000), so these videos easily add up to many hours of material. However, the organization and edition of the large amount of information contained in home videos present technical challenges due to the lack of efficient tools. Though a number of prototype systems for content-based video analysis and retrieval have been constructed, for example, as shown in (Wactlar, 1996; Chang, 1998), the development of tools and systems specialized for addressing home video, that is for extracting, representing, organizing, browsing, querying and retrieving video, is just on a preliminary stage (Huang, 2005; Wu, 2005). Several tasks are needed to confront to make the organization of home video possible and feasible. Home video has certain particular characteristics. The organization of home video should be based on the understanding of video structures, and by taking advantages of this structure. Home video are completed and stored straight in compressed domain. In order to save both time and space, techniques that manipulate home videos directly in compressed domain should be considered (Wang, 2003). Some typical techniques working on compressed domain could be found in (Taskiran, 2004). Home video are made by shot after shot without storyline, these shots may or may not have immediate relationship. To group shots, the visual features should be extracted from every shot (Gatica-Perez, 2003) Facing these tasks and difficulties, a novel technique is described in this article. It is based on the analysis of characteristics of home video, on the detection of motion attention regions in compressed domain, on the time weighting based on camera motion, and on a novel two-layer shot clustering approach and organization strategy. Experiments made on two home videos from MPEG-7 data set provide encouraging results.