Image Segmentation Using Particle Swarm Optimization and Simulated Annealing

A significant amount of picture segmentation is necessary for the accurate identification and diagnosis of the object of interest present in the image. Low image contrast and poorly defined boundaries are the problems of automatic segmentation. While there are many other picture segmentation techniques available, classic approaches are still popular despite having significant limitations that prevent them from producing reliable results. The main objective of this paper is to segment objects in an image by applying combined optimization techniques namely Particle Swarm Optimization (PSO) and Simulated Annealing (SA). PSO's worldwide exploration capability and SA's local refinement capability were combined for more effective segmentation. We also investigated, how the change of cluster size affects the output and the quality metrics. An increase in cluster size does not increase the quality metrics, instead, segmentation accuracy is increased based on the proper selection of cluster size. PSO-SA method is applied to different datasets, and segmentation accuracy is measured using several metrics namely precision, recall, F1 score, dice, and jaccard similarity.