Multithresholding of Benchmark Images by A Novel Optimization Approach

Abstract

Optimization based multithresholding techniques operates a cost function in order to segment an image via the obtained threshold values. For better segmentation results, a satisfier cost function and a robust optimization algorithm that is compatible with the used cost function, are needed. In this study, Scout particle swarm optimization (ScPSO) containing the efficient parts of Particle Swarm Optimization (PSO) and Artificial Bee Colony Optimization (ABC) is chosen for the optimization based process. As being the cost function, Kapur is preferred according to the advices in literature. Thus, Kapur-ScPSO technique is formed for the task of image segmentation. For performance comparison, ScPSO is compared with PSO and Genetic Algorithm (GA) on segmentation of four well-known benchmarking images (Lena, Baboon, Hunter, Map). Standard deviations, objective values and Total Statistical Success (TSS) values are calculated for every algorithm at the evaluation of performances. All algorithms are employed 50 times to choose the best performance. Consequently, it's seen that Kapur-ScPSO achieves to better standard deviations and objective values than Kapur based PSO and GA algorithms on image segmentation. Furthermore, TSS values of proposed method are brilliant on both statistical metrics.