A New Approach in Cluster Analysis

Abstract

A new clustering approach that is capable of finding clusters that are separated by some complex hypersurface is proposed. The approach can be useful for performing analysis of big amounts of unlabeled images that can be nowadays easily gathered, in particular by using unmanned aerial vehicle with mounted cameras. The approach is based on “softening” the initial clustering criterion and then using nonlinear optimization to find the optimal hypersurface that separates clusters.