Shape-based image retrieval is achieved by measuring the spatial topology
distance between the input image and each template image in the database. The
contour of an object in the input image is regularly sampled as feature points
which are matched to those of a template image by a pseudo elastic matching
process. The elastic matching is achieved by performing the self-organizing map
algorithm on the network which is constructed by distributing neurons to
feature locations of the template image. The spatial topology distance is
measured by the degree of distortion of the template pattern before and after
elastic matching. We tested the method on the Columbia Object Image Library
database. Preliminary experiments suggested promising result by our approach.
