Epic: A Photograph Retrieval System Based on Evidence Combination Approach

In this paper, we describe a system which applies an evidence combination approach for the retrieval of photographs from large repositories. To achieve eeective retrieval we use features derived from various components (such as text, and image) of the photographic document. As photographic features we use spatial features, which are objects and their locations in the photograph. We use these spatial features along with other textual features derived from the various components of the document. A user can search based on any of the features or in combination. For the combination of evidence we apply the Dempster-Shafer theory. In this combination scheme, we achieve the integration of database and information retrieval mechanisms into one computational model. The architecture of the system is based on a client-server architecture. The server is built using ECLAIR, an extensi-ble class library for information retrieval developed on top of an object-oriented database management system. The client is built using the Java programming language. The architecture of the server allows the integration of DB and IR approaches at the implementation level. The usefulness of the spatial features is evaluated using practicing graphic designers and has given very promising results. 1 Introduction Due to the widespread use of photographs in everyday life, there is a need for storing, managing and retrieving photographs. A large amount of pictorial information is produced in domains like the newspaper industry, the advertising business and the publishing industry. Journalists may look for photographs that describe an event or concern a person. Advertising professionals and publishers may look for photographs that are most useful for conveying their ideas. A number of commercial as well as in-house photographic repositories support these large populations of de