Images information retrieval using Gustafson-Kessel relevance feedback

Abstract

The goal of CBIR is to retrieve images that are visually similar to the query image. Relevance feedback retrieval systems ask the user for feedback on retrieval results and then use this feedback on later retrievals with the goal of increasing retrieval performance. The objectives of this research are to compare CBIR based with Gustafson-Kessel (GK) clustering and relevant feedback approach and to evaluate the effectiveness of GK relevance feedback for images retrieval. The research requires a better understanding of GK clustering and probabilistic relevance feedback method in turn to figure out different methods that can be used in solving similar problems. This project will give better insights in the usage of relevance feedback learning in order to reduce the gap between low-level features and high-level human concepts. The research will evaluate Gustafson-Kessel clustering and probabilistic relevance feedback method to improve the retrieval performance.