Tremendous technological advances in the Internet era have enabled a much broader community to have the capability to access large digital libraries. Examples of content stored in such digital libraries include photographic images, scanned documents, satellite images, seismic data, medical images, and the ever increasing video data in which terabytes of data are continuously acquired and stored. Access and retrieval of such digital libraries are thus severe challenges for conventional access methods, which typically require the search and retrieval of annotations and metadata extracted from the content. Unfortunately, annotation and metadata extraction usually requires intensive human intervention and is fundamentally less scalable.
Indexing and retrieval of images and videos by means of content have received tremendous interest recently, as this technique offers strong hope of retrieving multimedia content through nontraditional access methods. Some of these technologies have already found their way into Internet search engines (for example, AltaVista Photo and Media Finder), while others have become commercially available products (for example, IBM QBIC, Virage, and Islip). There is also an emerging consensus that content description at both feature and semantic levels will be widely used as the infrastructures for distributing multimedia content become mature. As a result, there is an intensive ongoing effort in the MPEG-7 community to define a set of descriptors, description schemes, and data description languages to ensure interoperability of content descriptions among different servers and clients.
Computer vision and pattern recognition play an important role in solving many critical problems in this area, in conjunction with other disciplines (e.g., image processing, databases, and information retrieval). On June 21, 1998, we organized the IEEE Content-Based Access of Image and Video Library Workshop (CBAIVL'98), colocated with the annual Computer Vision and Pattern Recognition (CVPR) Conference, to discuss these topics. This special issue is an outgrowth of the workshop.
As we outlined in the original call for papers, we think that the most important issues in content-based access of both image and video libraries are genuine application scenarios with a sound performance evaluation metric. These two issues are really closely associated with each other, as only genuine applications can be used to derive benchmark queries from which performance can be evaluated. Iterative refinement and learning is another important area, as there is no universal similarity measure that can be applied. Consequently, intensive user interactions are usually required to produce reasonable results. How to adapt the system to these user interactions is thus of great interest. Additional issues, such as feature extractions, classifications, analysis and retrieval techniques, and query framework, are also essential issues to produce a successful system that supports content-based queries.
In addition to the merits of each paper, the final selection of the 13 papers for this special issue strongly reflects out philosophy in this research area. Among these papers, four papers describe new approaches in feature extraction, classification, and segmentation; one paper addresses high-dimensional indexing, one paper addresses indexing in the transformed domain (specifically, the wavelet domain), one paper focuses on learning and relevance