Maximum entropy segmentation of improved particle swarm optimization

Abstract

The most common method of image segmentation is the maximum entropy threshold segmentation, which is simple to implement and easy to segment. The standard maximum entropy segmentation has the problems of slow operation speed and low efficiency. Therefore, particle swarm optimization (PSO) is used to optimize the optimal threshold vector of the maximum entropy threshold segmentation. However, the traditional particle swarm optimization algorithm have some shortcomings, such as fall into dimensional disaster and premature convergence. Therefore, a maximum entropy segmentation algorithm based on improved particle swarm optimization is proposed. Improved particle swarm optimization uses multiple one-dimensional combinations to replace the original multi-dimensional group. Information exchange between these one-dimensional particle swarms produces the overall fitness value of the particle swarm, and then replace the worst particles with the best particles in each one-dimensional group, thereby eliminating premature convergence. Finally, the method is applied to liver image segmentation and compared with the standard particle swarm maximum entropy threshold segmentation. The results show that this improvement has better threshold and faster convergence.