A Multi-level Thresholding Approach Based on Group Search Optimization Algorithm and Otsu

Abstract

Image Segmentation is a key process in image analysis and computer vision. Otsu is a simple but effective thresholding method, which is widely used for image segmentation. However, when one-dimensional Otsu is generalized to multi-threshold, the increased amount of computation will break down its efficiency and limits its application. Some evolutionary algorithms haven utilized to speed up the basic multi-level Otsu, such as genetic algorithm, particle swarm optimization, differential evolution algorithm etc, but these algorithms are easy to trap into the local optima. In the paper, in order to reduce computation and obtain the optimal thresholding values, the group search optimizer (GSO) algorithm is employed to optimize the basic Otsu thresholding method. The presented approach has been tested on some standard images and compared with other evolutionary algorithms in terms of fitness value. Experimental results prove that GSO is robust and superior to the other methods involved in the paper.