Content-based Image Retrieval Content-based Image Retrieval

Sensing and processing multimedia information is one of the basic traits of human beings: The audiovisual system registers and transports surrounding images and sounds. This complex re-cording system, complemented by the senses of touch, taste, and smell, enables perception and provides humans with data for analysing and interpreting the environment. Imitating this perception and the simulation of the processing was and still is one of the major leitmotifs of multimedia technology developments. The goal is to find a representation for every type of knowledge, which makes the reception and processing of information as easy as possible. The need to process given information, deliver it, and explain it to a certain audience exists in nearly all areas of day-to-day life: commerce, science, education, and entertainment (Smeulders, Worring, Santini, Gupta, & Jain, 2000). The development of digital technologies and applications allowed the production of huge amounts of multimedia data. This information has to be systematically collected, registered, organised, and classified. Furthermore, search procedures, methods to formulate queries, and ways to visualise the results have to be provided. In early years, this task was tended to by existing database management systems (DBMS) with multimedia extensions. The basis for representing and modelling multimedia data is so-called binary large objects, which store images, video, and audio sequences without any formatting and analysis done by the system. Often, however, only a reference to the object is handled within the DBMS. For the utilisation of the stored multimedia data, user-defined functions (e.g., content analysis) access the actual data and integrate their results in the existing database. Hence, content-based retrieval becomes possible. A survey of existing retrieval systems was presented, for example, by Naphade & Huang (2002). This article provides an overview of the complex relations and interactions among the different aspects of a content-based retrieval system, whereby the scope is purposely limited to images. The main issues of the data description, similarity expression, and access are addressed and illustrated for an actual system. BACKGROUND