A New Dissimilarity Measure for Clustering with Application to Dermoscopic Images

This paper explores the use of a new dissimilarity measure for clustering image databases with a special focus on dermoscopic images. It considers the area of the largest square sub-matrices approximately matching in the two images for computing their dissimilarity. A variant of the K-medoids approach is proposed using the new dissimilarity measure in the optimisation function. An experiment compares the proposed clustering approach with other six well-known methods in terms of F-Normalised Mutual Information, Adjusted Rand Index, Jaccard index, and purity, on a dermoscopic image database characterised by 12 skin diseases. The obtained results show that the new clustering approach is very promising in clustering dermoscopic image databases versus the other competing approaches, which is a valid support in speeding up the medical diagnosis process.