Research On 2D-OTSU Image Segmentation Algorithm Based On Swarm Intelligence Optimization

Image segmentation technology is the most important part of digital image processing,and its efficiency and accuracy determine the efficiency of subsequent image processing.Due to the diversity and complexity of image formats and types,as well as some special target requirements for image presentation,the traditional threshold segmentation method has been unable to meet,and how to improve the efficiency and accuracy of image segmentation has become an urgent problem to be solved.Starting from solving the optimal threshold of 2D-Otsu image segmentation,this paper applies the improved sparrow search algorithm and crow search algorithm to2D-Otsu image segmentation.In this paper,four classical images are selected for image segmentation experiment,and the optimal threshold,fitness value,iterative convergence curve and two image quality evaluation indexes of each algorithm are compared.The results show that NSSSA and ICSA algorithms have achieved good results in 2D-Otsu image segmentation.The main work of this paper is as follows:(1)A Non-uniform spiral learning Search Sparrow Search Algorithm(NSSSA)was proposed to solve the problem of fast convergence speed but low convergence accuracy of sparrow search algorithm.The algorithm uses a new learning selection strategy to improve the learning ability of the algorithm during optimization,and then uses a non-uniform spiral strategy to balance the local development and global search ability of the algorithm.By using the above method,the effectiveness of the algorithm is verified on 6 benchmark functions,CEC2013 and CEC2017 test sets.Experiments show that the accuracy and speed of the algorithm are improved,and the PSNR and SSIM values of the image quality evaluation index after segmentation are better.(2)In order to avoid the Crow Search Algorithm falling into local optimization and improve its global optimization ability,a Improved Crow Search Algorithm(ICSA)is proposed in this paper.The improved algorithm introduces differential variation perturbation factor to accelerate the information exchange within the population,which improves the convergence speed and effectively guarantees the convergence ability.Furthermore,the dynamic scaling rule of model parameters is introduced to ensure that the algorithm has better population diversity in the early stage,which not only improves the global search performance of the algorithm,but also effectively ensures that the algorithm has strong local search ability in the late stage.Through the test of 6 reference functions and engineering design application,it is proved that the convergence speed and solving accuracy of the algorithm are effectively improved,and the PSNR and SSIM values of the segmentation quality evaluation index on the four images are improved.