Improvement Of Differential Evolution Algorithm And Its Application On Multi-threshold Image Segmentation

Image segmentation is a classical problem studied in the field of computer vision and plays a significant role in image processing and image recognition.The main objective of image segmentation is to partition an image into several regions with different similar features.Threshold segmentation is a common method for image segmentation,however,due to the large differences in attributes between images and as the number of thresholds increases,the time cost of image segmentation increases and so does computational effort.The accuracy of image segmentation is critical to the outcome of image analysis,and the threshold segmentation results depends on the selection of thresholds.Therefore,for multithreshold image segmentation,this thesis combines multi-threshold segmentation algorithms with intelligent optimisation algorithms to find the optimal combination of thresholds.To address these problems of large amount of computation,high time cost and low segmentation accuracy in multi-threshold image segmentation,this thesis improves the differential evolution algorithm to improve image segmentation accuracy and efficiency in multi-threshold image segmentation.The main work of this thesis is as follows.(1)In this thesis,an adaptive dual mutation differential evolution algorithm(AHB-DE)is proposed,which mainly improves the differential evolution algorithm from two aspects: Firstly,an opposite learning strategy is introduced in the initial population to select the better individuals among the initial population and the opposite population,so as to improve the quality of the initial population;secondly,an adaptive dual mutation strategy based on historical thresholds differential evolution algorithm is designed to provide individuals with different mutation strategies to balance the exploration and exploitation ability of the algorithm.This strategy takes the number of successfully evolved individuals in the last generation as the threshold,and adaptively adjusts the superior and inferior individuals’ evolutionary direction.The experimental results show that the AHB-DE has better performance in improving the image segmentation accuracy and accelerating the convergence speed.(2)In this thesis,an improved algorithm named basic sequential clusteringbased differential evolution algorithm(BSC-DE)is proposed.Firstly,a local sequential population expansion strategy is proposed during population initialisation and applied to expand individuals in half of the initial population.This strategy expands new neighbourhood individuals near the better individuals,improving the quality of the initial population and allowing the algorithm to locate regions with better solutions early.Secondly,a sequential clustering mechanism is used in the evolution of the population to divide the niches,so that the population is assigned to co-evolve in different niches.On this basis,a mutation strategy based on central sampling is used to generate trial individuals for the initial population.Finally,the improved algorithm is experimentally tested on the function test set as well as on multi-threshold image segmentation.In comparison with other algorithms,the BSC-DE algorithm can better balance the development and exploration capabilities of the algorithm,and the effectiveness and competitiveness of the BSC-DE algorithm on the multi-threshold image segmentation problem is also verified.