With advances in computer technology, images and image databases are becoming increasingly important. Retrievals of images in current image database systems have been designed using keyword searches. These carefully designed and handcrafted systems are very efficient given the application domain they are built for. Unfortunately, they are not adaptable to other domains, not expandable for other uses of the existing information and are not very forgiving to their users. The appearance of full-text search provides for a more general search given textual documents. However, pictorial images contain a vast amount of information that is difficult to catalog in a general way. Further this classification needs to be dynamic providing for flexible searching capability. The searching should allow for more than a pre-programmed set of search parameters, as exact searches make the image database quite useless for a search that was not designed into the original database. Further the incorporation of knowledge along with the images is difficult. Development of an image knowledge base along with content-based retrieval techniques is the focus of this thesis. Using an artificial intelligence technique called case-based reasoning, images can be retrieved with a degree of flexibility. Each image would be classified by user entered attributes about the image called descriptors. These descriptors would also have a "degree-of-importance" parameter. This parameter would indicate the relative importance or certainty of that descriptor. These descriptors are collected as the "case" for the image and stored in "frames" Each image can vary as to the amount of attribute information they contain. Retrieval of an image from the knowledge base begins with the entry of new descriptors for the desired image. Along with the descriptors are the degree-of-importance parameter. The degree-of-importance would indicate the requirement for the desired image to match that descriptor. Again, a variable number of descriptors can be entered. After all criteria are entered, the system will search for cases that have any level of matching. The system will use the degree-of-importance both in the knowledge base about the candidate image(s) and the degree-of-importance on the search criteria to order the images. The ordering process will use weighted summations to present a relatively small list of candidate images. To demonstrate and validate the concepts outlined, a prototype of the system has been developed. This prototype includes the primary architectural components of a potentially real product. Architectural areas addressed are: the storage of the knowledge, storage and access to a large number of high-resolution images, means of searching or interrogating the knowledge base, and the actual display of images. The prototype is called the "Smart Photo Album" It is an electronic filing system for 35mm pictures taken by the average photographer on up to the photo-journalist. It allows for multiple ways of indexing the pictures of any subject matter. Retrieval from the knowledge base provides relative matches to the given search criteria. Although this application is relatively simple, the basis of the system can be easily extended to include a more sophisticated knowledge base and reasoning process as, for example, would be used for a medical diagnostic application in the field of dermatology.
Library of Congress Subject Headings
Expert systems (Computer science); Information storage and retrieval systems; Image processing
Department, Program, or Center
Computer Science (GCCIS)
Janicki, James, "Retrieval from an image knowledge base" (1993). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus