meta-heuristic algorithms
Forough Shahabi; Fereshte Pourahangarian; Homayoon Beheshti
Abstract
One of the fundamental problems in image processing is image segmentation identifying the objects and other structures in the image. Image thresholding is one of the widely used methods for image segmentation that can separate pixels based on the specified thresholds. The Otsu method calculates the thresholds ...
Read More
One of the fundamental problems in image processing is image segmentation identifying the objects and other structures in the image. Image thresholding is one of the widely used methods for image segmentation that can separate pixels based on the specified thresholds. The Otsu method calculates the thresholds to divide two or multiple classes. Classes are based on between-class variance maximization and within-class variance minimization. However, increasing the number of thresholds surges the computational time of the segmentation. To overcome this drawback, the combination of Otsu and the evolutionary algorithm is often effective. In this paper, we present a hybrid method utilizing the CSA and Otsu for multilevel thresholding. The result of our method has been compared with the three other evolutionary algorithms consisting of improved Particle Swarm Optimization (PSO), Firefly Algorithm (FA), and also the fuzzy version of FA. The evaluation consequence of the five benchmark images shows time and uniformity criteria have been improved.