A three-way clustering approach using image enhancement operations

Abstract

Three-way clustering receives its motivation from three-way decisions. It uses the core set and support set to describe a cluster. The two sets divide clustering results into three parts or regions called inside, outside, and partial. The division helps identify the central core and outer sparse regions of a cluster, which is useful when the clusters have dense regions but also have vague boundaries. One of the main challenges in three-way clustering is the meaningful construction of the two sets and three regions. In this article, we introduce a blurring and sharpening inspired three-way clustering algorithm or BS3 for short. We first explain the use of blurring and sharpening operations to create a three-way representation for a typical object in an image in the form of central primary (the clear part of the object), blurry (the unclear part of the object), and the non-object part. Next, by realizing similarities between the object and a cluster, we define cluster blur and cluster sharp operations to create a three-way representation for clusters. Experimental results on real-world and synthetic datasets suggest that BS3 is comparable to best performing approaches and in many cases has superior results.