Catálogo de publicaciones - libros

Video database applications require a suitable indexing technique to capture the time-varying nature of video data, together with a high performance retrieval strategy. This chapter presented an adaptive video indexing (AVI) technique and its integration with the specialized neural network model, which impressively satisfies these requirements. The template-frequency model (TFM) demonstrates very effective representation of the dynamic content of video, at various levels, and offers retrieval-by-video-clip to facilitate retrieving video groups and stories. Since TFM model addresses the difficulty in capturing spatio-temporal information, it allows relevance feedback analysis to capture information on the dynamic content, rather than spatial information offered by key-frame based modeling techniques. Unlike previous video database search attempts, we incorporated a self-learning neural network to implement an automatic relevance feedback method, which requires no user input for its adaptation. Based on a simulation study, this adaptive system, utilizing the TFM and automatic-RF retrieval architecture, can be effectively applied to a video database, with promising results. This approach combines many new features, which may help to usher in a new generation of video database applications.

This chapter presents applications of adaptive video indexing (AVI) technique for movie retrieval. AVI is demonstrated by an online multimedia search engine for retrieval using the query-by-video clip method, and compared to other popular video indexing and retrieval techniques. The search engine implements AVI in both user-controlled and self-organizing relevance feedback processes, which provides very high accuracy retrieval results as well as quick query response time. The chapter also presents an integration of AVI with an audio-visual fusion model that can support high-level concept queries. This was successfully applied for retrieval of movie clips from a large digital collection.