Analysis of Similarity Measurements in CBIR Using Clustered Tamura Features for Biomedical Images

Content based image retrieval (CBIR) is an important research topic in many applications, in particular in the biomedical field. In this domain, the CBIR has the aim of helping to improve the diagnosis, retrieving images of patients for which a diagnosis has already been made, similar to the current image. The main issue of CBIR is the selection of the visual contents (feature descriptors) of the images to be extracted for a correct image retrieval. The second issue is the choice of the similarity measurement to use to compare the feature descriptors of the query image to ones of the other images of the database. This paper focuses on a comparison among different similarity measurements in CBIR, with particular interest to a biomedical images database. The adopted technique for CBIR is based on clustered Tamura features. The selected similarity measurements are used both to evaluate the adopted technique for CBIR and to estimate the stability of the results. A comparison with some methods in literature has been carried out, showing the best results for the proposed technique.