Abstract

Abstract: RelevanceFeedback (RF) approach refines the retrieval process and thereby improves theperformance of CBIR system. Related works on CBIR are investigated and it wasobserved that existing Relevance Feedback techniques face the challenges ofnumber of iterations and the execution time. To improve the retrievalefficiency of the existing system, the proposed RF approach makes use of binaryclassifier and a feature selection technique to reduce the dimensionality ofthe image feature space. The positive and negative examples provided by theuser will be used to determine a small number of the most important featuresfor the classification in every RF iteration. The trained classifier will belater used to provide an updated ranking of the database images represented inthe space of the selected features.

Keywords:  Content based image retrieval (CBIR), RelevanceFeedback (RF), Iterations, Classification, Ranking.