Computer Vision and Multimedia Information Systems

An image database system with content capturing and fast image indexing abilities " , Proc. Srinivasan, " On-line images from the history of medicine (OLI): Creating a large searchable image database for distribution via WorldWide Web " , First [48] A. Rosenfeld, " Recognizing unexpected objects: a proposed approach " , Int. A construction of a new image database system which realizes fully automated keyword extraction, IEICE Trans. 8 In the context of MIS, the relevance and attentional concepts can be used for images archival, images retrieval , and for human-media interaction. For images ar-chival, these concepts allow to select important features, to rank them according to their pertinence, and to locate items of interest in the images. It is then possible to use these most important items as indexing keys for access-ing pictures. Regarding images retrieval, relevance and attentional mechanisms appear as general paradigms for exploratory data mining, that provide a framework for qualitative indexing and matching which can be used for QVE and possibly for QSD. Finally, regarding human media interaction, relevance could be used in two ways. First, at the input of complex conjunctive queries, users could be asked to provide a relevance factor together with each component of their request. Second, for presenting results to the user, relevance could be used to rank the images retrieved from the database with respect to the query. The results of the proposed technique have been described in the context of an artificial, still complex image database. Our current work consists of the integration of these concepts into an images archiving and retrieval system, applied to news photographs and to textile samples. We are also investigating learning techniques for automatic construction of object descriptions, aiming at precompiling appropriate indexing keys for efficient image retrieval. To conclude, computer vision offers a number of techniques that can be used in the context of pictorial information systems. More specifically, queries by means of visual examples need to heavily rely on pattern recognition and object matching methods. Future challenges are in the domain of automated determination of indexing keys from large data sets, and in the development of new interaction models that integrate computer vision methods.