Improving the estimation of distribution algorithm with a differential mutation for multilevel thresholding image segmentation

Image segmentation consists of separating an image into regions that are entirely different from each other, and multilevel thresholding is a method used to perform this task. This article proposes an Estimation of Distribution Algorithms (EDA) combined with a Differential Evolution (DE) operator as a metaheuristic to solve the multilevel thresholding problem. The proposal is called the Differential Mutation Estimation of Distribution Algorithm (DMEDA), where the inclusion of the Differential Mutation increases the standard EDA’s exploration capacity. The performance of the DMEDA for image segmentation is tested using Otsu’s between-class variance and Kapur’s entropy as objective functions applied separately over the Berkeley Segmentation Data Set 300 (BSDS300). Besides, a comparative study includes eight well-known algorithms in the literature. In this sense, statistical and non-parametric tests are performed to verify the efficiency of the DMEDA in solving the image segmentation problem from an optimization perspective. In terms of segmentation, different metrics are employed to verify the capabilities of the DMEDA to segment digital images properly. Regarding the two objective functions, the proposed DMEDA obtains better results in 97% of the experiments for Otsu’s between-class variance and 85% for Kapur’s entropy.