Employing Region Features for Searching an Image Database

This paper describes recent work on the use of regional data extracted from segmented images for use as search keys in an image database query system. The motivation for this work is the increasing availability and use of digital photography within the home and the lack of suitable search facilities based on the content of the images therein. The selection of images from the database is performed by Radial Basis Function networks, the placement of which is carried out by Kohonen's Learning Vector Quantisation approach. The work, which has been carried out in collaboration with Hewlett Packard Labs in Bristol, has shown that the use of features extracted on a local regional level, rather than a global scale can provide a high success rate when used as keys for a database query. A manual region classi cation was carried out on a test set of images in order to evaluate the performance of the system and it was found that, of the regions returned with a classi cation, an impressive 76.6% matched that used for the search key.