Research On Multi-threshold Image Segmentation Method Based On Swarm Intelligence

The main work of image segmentation is to divide an image into several regions with specific properties and extract the objective of interests.Among all the existing techniques,multi-level thresholding methods are widely used in multi-objective image segmentation whereas its simplicity and directness.However,it has a large amount of calculation and high complexity,which seriously affects the efficiency of segmentation.The key of the multi-level thresholding methods is to quickly find optimal thresholds in complex parameter space with certain criteria.While dealing with complicated nonlinear and multi-dimensional problems,the swarm intelligence optimization algorithms can find a satisfactory global optimal solution and reduce calculation time.Hence,to improve the efficiency of multi-level thresholding methods,swarm intelligence algorithms for multi-level image thresholding have become a research hotspot.However,these methods generally have a slow search speed and are easy to fall into local optimum during the later period which causes low precision.Focusing on traditional cuckoo search algorithm,particle swarm optimization and whale optimization algorithm in this study,to improve the precision and speed of segmentation,different strategies are proposed and involved to search multi-level thresholds.The main research results are summarized as follows:(1)To improve the traditional cuckoo search algorithm,the step size of cuckoo search algorithm is made adaptive from the fitness value of the current population.Cuckoo search algorithm has been applied to multi-level thresholding.However,with the number of thresholds increasing,the amount of computation increases exponentially.The main reason is that it uses a fixed step size and ignores the individual optimization capabilities.Therefore,the step size in the Levy flight is adaptively decided by its fitness value of the current population in this study,and the constant discovery probability pa in the biased random walk is changed automatically with respect to the current and total iterations.The population diversity and the capability of jump out of the local optimum will be enhanced by these improved strategies during the later period,and avoid premature.To verify segmentation accuracy and efficiency of the proposed method,an adaptive cuckoo search algorithm proposed by Naik and traditional cuckoo search algorithm are included to compare.The results show that the proposed algorithm is expert in selecting optimal thresholds quickly and the segmentation accuracy is significantly improved.(2)The chaotic perturbation strategy generated by Circle map is used to improve the Darwinian particle swarm optimization algorithm and applied to multi-level color image thresholding.Owing to the Darwinian particle swarm optimization algorithm is easy to trap into the local optimum during the later period,it leads to inaccurate segmentation results.Because of the randomness,ergodicity and initial value sensitivity of chaotic motion,hence the chaotic perturbation strategy generated by Circle map was applied to the global optimal particle position,which could effectively prevent the algorithm from falling into local optimum,improve convergence speed and search accuracy.The chaotic Darwinian particle swarm optimization algorithm is used to optimize Kapur's entropy which considered as the objective function,the color image can be segmented by the obtained optimal thresholds in this study,to illustrate the advantages and disadvantages of the proposed algorithm,three color images are selected,the multi-level thresholding results obtained by chaotic Darwinian particle swarm optimization algorithm are compared to Darwinian particle swarm optimization algorithm,particle swarm optimization algorithm and harmony search algorithm.The experimental results show that the introduced method extracts the required targets effectively and improves the segmentation quality of color images.(3)Using McCulloch's method can efficiently generate stable random variables to improve the whale optimization algorithm,and applied to multi-level remote sensing image thresholding.The traditional whale optimization algorithm has a low precision in the later period,resulting in poor image segmentation.Therefore,the McCulloch's method is applied to perturb the current optimal whale position,it can make a fine search near the optimal location and expand the search scope,increase the diversity of the population and balance the global search and local search,and then,improve the accuracy of the algorithm and avoid premature.With Otsu's method as the optimized objective function,whale optimization algorithm is exploited to maximize the Otsu's method to obtain the optimal thresholds and extract the required target from the image accurately.To verify the accuracy of the proposed algorithm,three remote sensing images are selected,the improved whale optimization algorithm are compared with other methods,the experimental results show that the improved whale optimization algorithm has higher accuracy and improves the segmentation quality of remote sensing images.Figure[35]Table[13]Reference[122]...