Estudo de algoritmos de otimização bio-inspirados aplica à segmentação de imagens
Image segmentation is one of the first steps within the framework for processing scenes. Among the main existing techniques, we highlight the histogram-based binarization, which due to the simplicity of understanding and low computational complexity is one of the most used methods. However, for a multi-threshold process, this method becomes computationally costly. To minimize this problem, optimization algorithms are used to find the best thresholds. Recently, several algorithms inspired by nature have been proposed in a generic way in the area of combinatorial optimization and obtained excellent results, among which we highlight the more traditional ones such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Differential Evolution Algorithm (DE), Artificial Bee Colony (ABC), Firefly Algorithm (FA) and Krill Herd (KH). This work shows a comparison between some of these algorithms and more recent algorithms, from 2014, as Grey Wolf Optimizer (GWO), Elephant Herding Optimization (EHO), Whale Optimization Algorithm (WOA), Grasshopper Optimization Algorithm (GOA) and Harris Hawks Optmization (HHO) . This work compared the thresholds obtained by 7 bio-inspired algorithms in a base composed of 100 images with 1 single object provided by the Weizmann Institute of Science (WIS). The comparison was made using consolidated metrics like Dice/Jaccard and PSNR, as well the recent Hxyz. In the experiments were used the extensive system as an objective function (Kapurs´ Method). Still in the proposal of this experiment, the extensive system was compared with a Tsallis nonadditive entropy, with the Super-extensive system being configured with q ? [0.1, 0.2, . . . 0.9] and the Sub-extensive system with q ? [1.1, 1.2, . . . 1.9]. The image Database contains 100 images with only 1 object on scene. The results show that the Krill Herd (KH) algorithm was the winning algorithm in 35% of executions according to the PSNR metric, 28% in the Dice/Jaccard metric and 35% on the Hxyz metric. The extensive system had the best overall performance and was responsible for the best threshold of 54 images according to the metric PSNR, 30 according to the metric Dice/Jaccard and 39 according to the Hxyz metric