Refined neutrosophic sets in content-based image retrieval application

This paper presents a short contribution for content-based image retrieval (CBIR) systems using refined neutrosophic sets. First, images are represented with dominant color descriptor (DCD) which is an efficient tool for compact color representation. Then, each dominant color and its corresponding partition in DCD is considered as an object in image. Objects are modeled as refined neutrosophic subsets. With respect to CBIR application, new definitions of true (T) and indeterminacy (I) subsets are presented in refined neutrosophic domain which address some challenges in DCD. This is done by considering spatial information of pixels in DCD partitioning and refining wrong labels assigned to pixels. A new operation in refined neutrosophic space is also proposed. Furthermore, a new similarity measure is proposed for transformed images to refined neutrosophic domain. Benefits and weaknesses of the proposed method are explained in discussion section. Experiments show the improvement of at least 4% with respect to precision measure in comparison with other CBIR methods.