The Study Of Image Segmentation Based On Differential Evolution Particle Swarm Optimization And Fuzzy Clustering

Image segmentation is one of the basic techniques of image processing. Because of theambiguity of the image, the image segmentation algorithm based on fuzzy clustering graduallyget people’s attention. Fuzzy C-Means clustering algorithm (FCM) is one of the most completeand the most commonly used fuzzy clustering method, which uses non-linear programmingapproach to traditional clustering into a mathematical optimization problem with constraints andusing gradient iterative way to get optimal cluster center value. Image segmentation based onFCM using an unsupervised learning method for image segmentation, image data can be achievedself-learning and auto-segmentation, it reduces human intervention. However, FCM itselfaffected by the initial cluster centers and it will fall into local optimum.With the development of intelligent computing, more and more intelligent optimizationalgorithm are applied to image segmentation. Differential evolution algorithm (DE) and particleswarm optimization (PSO) are two global optimization algorithms. However, the two algorithmshave limitations, this article attempts to integrate the two algorithms FCM algorithm to optimizeand achieve image segmentation. This paper studies the FCM image segmentation, DE and PSOintelligent optimization algorithms, the main work and innovation can be summarized as follows:(1) Summarize the research status, the main advantages and disadvantages, and the majorimprovements strategy of FCM image segmentation. Then introduce the origin of fuzzyclustering and DE and PSO, and their advantages and disadvantages, improvement strategies andapplications.(2) In order to solve the problem of easily fall into local optimum, a new image segmentationalgorithm called DEPSO-FCM is proposed, it improves the FCM algorithm by using DE andPSO. This algorithm incorporates the global search ability of DE and the local search capabilityof PSO. On the one hand, it accelerates the DE’s late convergence rate; on the other hand, itoptimizes the PSO’s precocious. Take advantages of the two algorithms to optimize FCM imagesegmentation, and obtain the best optimization. Experiments show that the method has goodglobal search ability and convergence, for grayscale images with fuzzy noise, it can also get goodsegmentation results, its noise immunity is better than FCM.(3) Introduce and compare several widely used color space (include RGB, HSV and HSI),and make use of the DEPSO-FCM to segment image in different color spaces. Experimentalresults show that the algorithm can be divided images in different types of color space. Inparticular, it can also split the color images that contain the singular value points. When dividing the H component, it can effectively avoid the FCM algorithm into local extreme and get bettersegmentation results. The DEPSO-FCM has better stability and robustness.