Proto-reduct Fusion Based Relevance Feedback in CBIR

Abstract

This paper proposes two related RF methods for use in CBIR. These two methods are based on a general classificatory analysis based framework for RF in CBMR that considers RF independently from retrieval. The proposed methods show how the user's information need expressed as a set of "proto-reducts'' can be used as the basis of a re-weighting technique that can improve subsequent retrieval The performance of the proposed methods is studied on two image collections with different characteristics and compared against an existing RF method.