Searching large image databases using radial-basis function neural networks

This paper has demonstrated the effectiveness of the approach of using region-based feature data as keys to an image database query search. The classification engine required for this form of query has been implemented using a radial-basis function (RBF) network and it has been shown that this can provide significant results using low volume training data provided in the form of feedback from the user on the progress of the query. Having proven the feasibility of this technique, it is hoped that improvements to the feature set will result in higher system performance. It is also envisaged that extra tools will be added to the query system such as the Boolean combination of regions-of-interest for more detailed and customised query keys. It is also envisaged that hierarchical class structures can be built up containing super-and sub-classes. For instance, a super-class containing all examples of buildings may exist, as well as a sub-class of churches. The manner in which this form of structure will be created will depend on the user´s requirements