A NOVEL CBIR TECHNIQUE FOR RETRIEVAL OF SIMILAR MR IMAGES

Abstract

We investigate a new approach for content-based image retrieval used for accessing similar MRI scan images. The interesting aspect of this approach include a hierarchical combination of segmented centroid computation based Star-like graph and fuzzy feature matching approach. Large number of images, generated during investigation procedure is time prohibitive for comparative study of similar ailment cases manually. The task of accessing similar images accurately is really challenging. In the proposed system, similarity matching process consists of two steps in order to reduce the image retrieval time. In first step, algorithm scans through the entire database by comparing the image salient features for inexact similarity matching. The images with similarity measure closest to the query image are selected for further comparison. In the second stage we use a reliable region based fuzzy feature-matching approach to identify the images those are very similar to that of query image and indicated by similarity index. It is found to give better average precision, average recall and quicker retrievals than application of fuzzy feature matching (or graph matching) algorithm alone as shown with the help of Matlab Simulation. It can be used as a tool to physicians for diagnosis, tor surgeon for planning and to medical students. Key words: CBIR - Content-based image retrieval, Similarity measure. Fuzzy Feature Matching, MRI –Magnetic Resonance Imaging, fMRI – Functional Magnetic Resonance Imaging.

View PDF