Trial Realization of Human-Centered Multimedia Navigation for Video Retrieval

A trial realization of human-centered navigation for video retrieval is presented in this paper. This system consists of the following functions: (i) multi-modal analysis for collaborative use of multimedia data, (ii) preference extraction for the system to adapt to users' individual demands, and (iii) adaptive visualization for users to be guided to their desired contents. By using these functions, users can find their desired video contents more quickly and accurately than with the conventional retrieval schemes since our system can provide new pathways to the desired contents. Experimental results verify the effectiveness of the proposed system. was reported that the digital universe in 2007 was 281 exabytes and that it will be 44-times larger in 2020 than that in 2009. Also, most of the data contained in the digital universe are unstructured data such as images, music and videos, i.e., multimedia data. Nowadays, users always access the data to find useful information, generally based on the query-response model, in which users input a query into the retrieval interface and then get the desired information. The query-response model can narrow huge information down to a suitable size for users to check according to the query. Actually, metadata are attached to multimedia contents to be retrieved in advance; retrieval engines present 2 multimedia content, the metadata of which are matched to the queries, to users; after users input queries once or more than once, retrieval engines effectively guide users to the useful information. Traditionally, metadata of the contents were generated from information such as dates, times, and places, which were obtained in their acquisition; and manually annotated keywords were also utilized as metadata. Recently, image analysis and video analysis have been improved, and technologies for automatically extracting metadata have therefore been proposed (Lowe, Object development of these technologies, especially for deriving semantic level indices, and it is accelerating the progress contents desired by users, semantic retrieval becomes feasible. However, in order to realize accurate estimation of semantic level indices, a sufficient amount of training data must be provided, and it becomes difficult to perform accurate estimation when a large amount of training data cannot be prepared. Generally, it is difficult to prepare enough training contents for each semantic level index, and the estimation of semantic level indices therefore has some errors. In such cases, since the semantic features obtained from the estimated semantic level indices also have errors, the distances between two …