An image retrieval system for distributed environments

Abstract

We describe a prototype image retrieval system for distributed environments based on 2D projection interval relationships (2D-PIR). The system transforms an image to a symbolic representation that describes spatial relationships among objects in the image. The image database consists of three parts: the real image repository (which is distributed) and the symbolic and iconic image repositories (which are stored locally). We describe the result of experiments on a small image database which compare the performance of our system against a keyword-based retrieval system and also against other content-based image retrieval systems.