Research Of Image Segmentation Algorithm Based On Improved FCM

Image segmentation is a process of segmenting an image into several distinct and non-overlapping regions according to different applications,and then and then separating the region of interest from the background.Due to the diversity and complexity of images,it has always been one of the most challenging tasks in image understanding and computer vision.Although many methods of image segmentation have been proposed,none of them is robust and effective for a large number of different images.In the existing image segmentation technology,clustering is one of the most common methods,because it is easy to implement and the algorithm complexity is low.The purpose of clustering is to divide a set into several clusters so that members of the same cluster are similar while members of different clusters are different.Clustering methods are generally divided into hierarchical method,graph theory method,density function decomposition method and objective function minimization method.Among them,the fuzzy c-means(FCM)clustering algorithm,as an unsupervised clustering algorithm,has been well applied in the research fields of image segmentation and data processing.This algorithm introduces the fuzzy concept into the membership degree of image pixels.Compared with the hard clustering k-means algorithm based on partition,it can retain more original image information in the image processing process,which has attracted extensive attention from researchers.It was first proposed by Dunn and popularized by Bezdek.However,the traditional FCM clustering algorithm does not fully consider the spatial information of the image when performing image segmentation,which makes the algorithm more sensitive to noise.At the same time,this algorithm needs to set the clustering center,number of classification and other parameters in advance,which is easy to fall into local optimization.Therefore,this paper improves the image segmentation effect from two aspects.One of the improvement measures is to overcome local optimization and improve the robustness to noise by genetically optimizing the improved FCM algorithm that adds spatial neighborhood information.Another improvement measure is to improve the algorithm's anti-noise and the ability to maintain the image details by using the spatial correlation of markov random field combined with the image neighborhood spatial information and gray information.The main work of this paper is as follows:This paper firstly analyzes the standard FCM algorithm completely and systematically,and introduces the characteristics of fuzzy clustering algorithm.Then,from the perspective of principle analysis,analyzing and interpreting the improved FCM algorithm with high attention in the field of image segmentation in recent years,and expounding the advantages and disadvantages of each algorithm.FCM algorithm is easy to fall into local optimal and is sensitive to noise.Thus,an image segmentation algorithm based on genetic Fuzzy C-Means clustering is proposed,using genetic algorithm's global optimization ability to overcome the problem of falling into local optimal value.It improves the robustness of image noise through adding the neighborhood spatial information to FCM objective function to constrain the membership function,so that the results of segmentation are more in line with expectation.The improved FCM algorithm based on spatial neighborhood information still has room for improvement in noise immunity and detail retention.We puts forward a improved FCM algorithm based on markov random field,combining with neighborhood information given two innovation points:(1)Combining the neighborhood spatial information with the gray information,a new prior membership degree estimation is proposed,which makes the algorithm complexity less,robust and effective.(2)The similarity measurement combined with the relevant information of the neighborhood was proposed,and the influence of the neighborhood on the central pixel was adaptively controlled through the value of prior membership degree,so as to improve the anti-noise performance of the algorithm.Simulation analysis in MATLAB environment,FCM algorithm,FCM_S algorithm,FCM_S1/S2 algorithm,EnFCM algorithm,FGFCM algorithm,FLICM algorithm,KWFLICM algorithm and ours algorithm are compared through simulation experiment.the classical segmentation image,the four-class synthetic image,the Berkeley database images and brain image in different simulated noise are segmented.The experimental results show that the improved algorithm effectively improves the robustness to noise and the ability to maintain image detail features.