Region-based image retrieval using the semantic cluster matrix and adaptive learning

Abstract

An image retrieval system, which reduces the semantic gap between the low level image features and the high level semantic concepts, using the semantic cluster matrix (SCM) and adaptive learning during testing, is proposed. This mechanism retrieves semantically relevant images for both trained and untrained image categories. The SCM groups the new categories into semantic clusters, records the cluster's semantic information, gets the relevance feedback (RF) from the user, and records this in the SCM. Thus, the SCM adaptively learns about the new categories during the testing time, and is able to retrieve semantically relevant images for untrained categories also. Experiments were conducted using the Caltech image dataset consisting of 101 categories of images. The obtained results demonstrate that the proposed approach achieves good performance in terms of retrieval accuracy.