Pattern recognition in information systems

The field of computing has changed significantly in the last decade. A major factor for this change is the evolution of the computing technology itself. Advances in signal processing hardware, powerful image rendering mechanisms, and high bandwidth communication facilities, have spurred an increased interest in various kinds of database management systems for nontextual information. These include multimedia, spatial and object oriented databases, and the inter-linkage of database management and artificial intelligence techniques. Interest in digital images, in particular, has increased enormously over the last few years, motivated at least in part by the rapid growth of imaging on the World-Wide Web (Eakins and Graham, 1999). A picture is worth a 1,000 words, says the old adage. Modern data systems, in areas ranging from surveillance, space exploration and medical imaging, accrue and store massive numbers of images for future use. The accumulated images, however significant, are of little value if they cannot be quickly retrieved. Digital image databases typically have been organized with human-assigned textual labels, a time-consuming and labor intensive process. Content-based image retrieval (CBIR), a technique for retrieving images on the basis of automatically-derived features such as color, texture and shape, is becoming increasingly important in visual information management systems (VIMS). Content-based retrieval analyzes image features to automatically identify image content, and hence can provide a more accurate and efficient basis for image query systems. Management of images within a large database is an important and emerging area of research. The computer vision community has focused on the design of image database management systems (IDMS). Efficient use of VIMS requires techniques that are quite different from conventional textual database management systems (Jain and Gupta, 1996). Work in this direction is still in development, and general solutions to key problems relating to regions of interest, segmentation, and retrieval are still being sought. Color and texture are two of the features that have traditionally been used to approach this challenging problem. At The University of Texas at Austin, we have found that image structure, derived by perceptual grouping, is a valuable tool in our quest for more efficient content-based image retrieval (Iqbal and Aggarwal, 2002c; Iqbaland Aggarwal, 2002b). This presentation highlights the use of structure, derived via perceptual grouping, for mage classification and retrieval. Our use of structure does not require image segmentation. A hands-on comparison of results using color, texture and structure to retrieve images containing both natural and manmade objects will demonstrate that collectively structure, color and texture form an excellent feature set for image retrieval (Iqbal and Aggarwal, 2002a). Our system CIRES: Content-based Image Retrieval System, available on the web (CIRES), retrieves images ranging from scenes of purely natural objects such as vegetation, water and sky, to images containing conspicuous manmade structures like towers, buildings and bridges. In addition, it incorporates the use of multiple query images, and relevance feedback from the user to further refine the search (Iqbal and Aggarwal, 2003).Future work includes developing methodologies for efficient indexing structures that will provide optimal resolution of similarity queries. Image and video searches typically involve high-dimensional feature spaces. For rapid retrieval it is not feasible to perform a linear search on a very large image collection. For this purpose we are developing methodologies to accelerate indexing and retrieval by using database management techniques, including analyzing hierarchical access methods using tree-type structures for feature representation. This will help in constructing an effective data structure for multimedia data repositories that facilitates efficient queries on complex objects. Future uses of our system in surveillance and video summarization will also be discussed.