Hierarchically searching the web for images

We have been developing approaches intended to aid searching of the World Wide Web for images. Initially these are based upon the contextual information that we can obtain from existing text-based search engines with the ultimate intention of applying the work to a distributed image database. Supplying a search engine with a phrase describing an image subject provides a set of potentially related documents, and contained within those documents a set of images in which the user may be interested. These images are then filtered to remove unsuitable examples and those remaining are presented to the user in an informative manner to enable rapid browsing, and if necessary, refinement. We present the initial findings of this work, along with some preliminary results and conclusions. We have experimented with methods for arranging the filtered images in a 2-dimensional space, using Sammon´s (1969) mapping and self-organising maps, based upon simple image features. An approach such as this has demonstrated some interesting concepts, and enabled us to produce an intuitive, hierarchical image browser