Multi-stage imaging retrieval: a relevance feedback retrieval with nearest neighbour clusters

Abstract

Imaging Retrieval is a retrieval strategy which is based on modal logic where documents are viewed as possible worlds which are related through an accessibility relation established using their similarity/dissimilarity coefficients. With these accessibility relation documents are grouped into clusters based on a nearest neighbour concept. The work reported in this paper sets out to implement and evaluate the imaging retrieval as a relevance feedback retrieval with nearest neighbour clusters. The retrieval is implemented in two variations: a one-stage and a multi-stage retrieval. The results obtained from the experiments are enough to show the viability and validity of this strategy and to support it as something worth looking into further.