Dynamic particle swarm optimization and K-means clustering algorithm for image segmentation

Abstract

K-means clustering is usually used in image segmentation due to its simplicity and rapidity. However, K-means is heavily dependent on the initial number of clusters and easily falls into local falls into local optimum. As a result, it is often difficult to obtain satisfactory visual effects. As an evolutionary computation technique, particle swarm optimization (PSO) has good global optimization capability. Combined with PSO, K-means clustering can enhance its global optimization capability. But PSO also has the shortcoming of easily falling into local optima. This study proposes a new image segmentation algorithm called dynamic particle swarm optimization and K-means clustering algorithm (DPSOK), which is based on dynamic particle swarm optimization (DPSO) and K-means clustering. The calculation methods of its inertia weight and learning factors have been improved to ensure DPSOK algorithm keeping an equilibrium optimization capability. Experimental results show that DPSOK algorithm can effectively improve the global search capability of K-means clustering. It has much better visual effect than K-means clustering in image segmentation. Compared with classic particle swarm optimization K-means clustering algorithm (PSOK), DPSOK algorithm has obvious superiority in improving image segmentation quality and efficiency.