Research On Color Image Segmentation Algorithm Based On FCM

Image segmentation is to divide the image into several parts according to certain characteristics and highlight the required target part.It is the key step and basic premise to complete complex tasks such as image processing,understanding and target recognition.The smallest unit of describing image information is pixel.The essence of image segmentation is to solve the clustering problem between similar pixels.Because of the imaging mechanism of an image and the noise interference in the process of transmission and reception,the image is fuzzy and difficult to segment the image information.Therefore,many researchers apply the idea of fuzzy clustering to image segmentation to solve this problem.Fuzzy c-means clustering algorithm(FCM)combines fuzzy theory and fuzzy analysis into pixel clustering,which can well distinguish the membership relationship of fuzzy characteristics of pixels.In this paper,based on FCM algorithm,color image segmentation is studied.Firstly,the related algorithm theory in the field of image segmentation is summarized.Secondly,the research trends and development trend of FCM algorithm in the field of image segmentation are analyzed in detail.Finally,two improved algorithms for color image segmentation are proposed aiming at two defects of traditional FCM algorithm,namely,poor noise resistance and high iteration complexity.The main research contents are as follows:The traditional FCM algorithm has poor anti noise performance,and it is directly applied to color image segmentation without considering the color,space and other feature information between pixels.In this paper,we propose to replace Euclidean distance with covariance Mahalanobis distance to measure the pixel similarity,and consider the spatial information of the neighborhood of the pixels to be clustered to improve the noise resistance.At the same time,the prior knowledge is introduced into the membership matrix,which makes the membership matrix correct and correct the wrong pixels in each iteration.Through the contrast experiment of anti noise,it is proved that the anti noise performance of the algorithm is greatly improved,and the segmentation effect is significantly improved.The traditional FCM algorithm has random initial clustering centers and high computational complexity.In view of this,this paper proposes to use SLIC algorithm to preprocess the color image to get the super pixel,and then according to the super pixel color,boundary and spatial position relations,the similarity weighted calculation is carried out to obtain the super pixel similarity comprehensive information.Finally,the super pixels are further divided into subsets with similar features by subtractive clustering method according to the comprehensive information of similarity,and the center is used as the initial clustering center of FCM clustering,and the super pixels are merged and clustered.The experimental results show that the speed of the proposed algorithm is significantly improved,and the segmentation effect is also significantly improved.