Segmentation of Magnetic Resonance Brain Images Through the Self-Adaptive Differential Evolution Algorithm and the Minimum Cross-Entropy Criterion

The segmentation is regarded as a vital step in preprocessing techniques for image analysis. Automatic segmentation of brain magnetic resonance images has been extensively investigated since with a precise segmentation can be identified and diagnosed several brain diseases. Thresholding is an important simple but efficient technique of image segmentation. Various strategies have been submitted to find optimal thresholds. Amongst those methods, the minimum cross-entropy (MCE) has been broadly implemented due to its simpleness. Although MCE is quite effective in bilevel thresholding, the computational cost increases exponentially the higher the number of thresholds (th) to find. This article introduces a new approach called MCE-SADE for multilevel thresholding using the Self-Adaptive Differential Evolution (SADE) algorithm. SADE is a robust metaheuristic algorithm (MA) that resolve general problems efficiently since, through evolution, the parameters and the proper learning strategy are continuously adjusted pursuant to prior knowledge. The optimum th values are found minimizing cross-entropy through SADE algorithm. The proposed method is tested in two groups of reference images; the primary group is formed of standard test images, while the following group consists of brain magnetic resonance images. In turn, MCE-SADE is compared with two metaheuristic algorithms, Grey Wolf Optimizer (GWO) and Competitive Imperialist Algorithm (ICA). From the experimental results, it is observed that MCE-SADE results improve in terms of consistency and quality in contrast to GWO and ICA based methods.