A Novel Hybrid Clustering Algorithm Based on Minimum Spanning Tree of Natural Core Points

Abstract

Clustering analysis has been widely used in pattern recognition, image processing, machine learning, and so on. It is a great challenge for most existing clustering algorithms to discover clusters with complex manifolds or great density variation. Most of the existing clustering needs manually set neighborhood parameter K to search the neighbor of each object. In this paper, we use natural neighbor to adaptively get the value of K and natural density of each object. Then, we define two novel concepts, natural core point and the distance between clusters to solve the complex manifold problem. On the basis of above-proposed concept, we propose a novel hybrid clustering algorithm that only needs one parameter M (the number of final clusters) based on minimum spanning tree of natural core points, called NCP. The experimental results on the synthetic dataset and real dataset show that the proposed algorithm is competitive with the state-of-the-art methods when discovering with the complex manifold or great density variation. INDEX TERMS Clustering, complex manifold, natural neighbor, natural core points, minimum spanning tree.

Minimax Bridgeness-Based Clustering for Hyperspectral Data

“…Using the core set Γ 50 allows to tackle two drawbacks of the minimax distance: sensitivity to noise and computational complexity. Core sets have previously been used for similar purposes [28,40,47]. To select a representative core set in a computationally efficient and parameter-free manner, we estimate the underlying probability density function of the data and discard the 50% least dense points.…”