| With the great progresses in the manufacture, storage and prevalence of multi-media information, multi-media and digital video becomes an indispensable part of our daily life. Explosive increase of multi-media resources requires for automatic analyzing, processing and classifying multi-media technologies eagerly. Therefore, multi-media processing and indexing techniques become one of the major topics in many research institutions.Limited by the technology itself and processing speed, it seems impossible to construct a generic video analysis/processing framework. Recent researches are mainly focused on a certain video types such as news, film, TV shows, sports video, etc. This paper proposes a novel shot boundary detection and key-frame extraction algorithm, and a anchor person detection technique as well. Both of them are based on uncompressed domain processing and promising detection results and computation speed has been archived.The proposed shot boundary detection algorithm combines both chrome and motion info of video. We utilize main color histogram to figure out the chromatic variance of color, and block motion metric of motion. Final result is gained by the combination of both. Key-frame plays an effective role in depicting shot content. Traditional key-frame extraction techniques are mainly based on in-shot frame clustering, which is not fit for stream media processing. Thus a subshot-based unsupervised clustering algorithm is proposed, which can be used for stream video processing. |
PDF Full Text Request