Fusion-based Multilevel Thresholding For Image Segmentation Using Evolutionary Algorithm

Image segmentation is one of the crucial steps involved in the process of digital image processing. Bi-level segmentation is a simple task, but it can only be used when there are only two distinct regions. However, more challenging real-world situations, such as medical images, have more than two distinct regions, which require multilevel segmentation. The histogram-based multilevel thresholding method is the most widely used as it is simple and fast. Most of the research work used one of the three fitness functions such as Otsu, Kapur's entropy, and Tsallis entropy. However, they don't provide good performance for different types of datasets. To overcome the drawback of the existing multilevel thresholding techniques, a new fusion-based multilevel segmentation technique is proposed. Six popular evolutionary algorithms are investigated to find the optimal threshold. Experiments are performed on two different publicly available MRI datasets to demonstrate the effectiveness of the proposed fusion-based multilevel thresholding for tumor segmentation. The proposed fusion-based segmentation technique, in combination with the BAT algorithm achieves better performance in comparison to existing multilevel thresholding methods in terms of Accuracy, F-measure, Jaccard Index, and Dice Score.