Thresholding Based Image Segmentation Using Intelligent Optimization Algorithms

Thresholding is a classical and effective technique for image segmentation.However,the selection of the optimal thresholds for different kinds of real-world images is a difficult problem.In addition,the request of real-time process and robustness in many applications also limits the performance of the thresholding technique.Recent years,with the rapid development of intelligent optimization algorithms,many researchers have focused their attention on using the intelligent optimization algorithm to improve the thresholding performance,where Particle Swarm Optimization(PSO),Cuckoo Search(CS)and Flower Pollination Algorithm(FPA)are three representative ones among them.In this essay,the principle of PSO,CS and FPA is firstly expatiated.Then,to improve the image thresholding performance,three improved algorithms are proposed respectively based on the discussion of the shortcomings of the three basic PSO,CS and FPA.The specific improvement measures are listed below.1)The improvement of PSO: a novel learning strategy is developed to improve the global convergence performance of PSO according to the local pollination strategy of FPA,as the basic PSO often suffers from the problem of premature convergence.In addition,a diversity enhancing strategy is proposed to improve the population diversity.The two strategy works together which can enhance the searching capacity and global convergence performance of the algorithm.2)The improvement of FPA: firstly,a neighborhood learning strategy is introduced to improve the local searching efficiency.Secondly,the exploration is enhanced according to the biological analysis on the background of the algorithm.Finally,a linearly decreasing parameter strategy is further proposed to keep the balance of the local searching and exploration.With the three measures,the overall searching capacity is improved and the balance of exploration and exploitation is also achieved.3)The improvement of CS: a fully informed strategy is introduced to improve the team cooperation and information exchange in the evolution of the population,which effectively improved the overall optimization performance.With the improvement measures above,the three proposed algorithms show obvious advantage over the corresponding basic algorithm on the mean objective function value,two image quality measures and the convergence performance,which strongly demonstrated the superiority of the proposed algorithms on the aspect of optimization performance and searching capacity.In addition,the stability and the visual segmentation performance are also improved according to the comparison on image segmentation results and the distribution of the thresholds obtained by different algorithms.