Instance-based Relevance Feedback using Cluster Density

G. Das and S. Ray (Australia)

Keywords

Content-Based Image Retrieval, Relevance Feedback, Instance-based, Cluster Density

Abstract

Relevance Feedback is a useful technique to reduce seman tic gap which is a bottleneck to a successful Content-Based Image Retrieval system. In this paper, we have discussed an Instance-based relevance feedback method that uses the closeness or similarity of an instance to a class of similar instances. We have achieved improvement in retrieval ac curacy by incorporating cluster density of relevant images in addition to the nearness of an image from the relevant and non-relevant image sets. We have experimented with four databases with total image numbers ranging from 1000 to 19511, encompassing both narrow domain and broad do main. With the cluster density approach, we have achieved an improvement of up to 3.7% in averaged precision value as compared to an existing method.

Important Links:



Go Back