Research On Image Segmentation Algorithm Based On MRF And Fuzzy Clustering

With the rapid development of artificial intelligence,the requirements for image acquisition and processing are increasing,which makes the image segmentation as the image preprocessing stage particularly important.The work of image segmentation is to divide the image into different regions according to different characteristics so as to separate the target from the background and prepare for the subsequent analysis.The collected digital images will produce different types of noise due to the influence of equipment and natural conditions.Therefore,the noise problem has become one of the difficulties of the image segmentation algorithm.Fuzzy clustering algorithm is one of the most widely used algorithms in many image segmentation algorithms.It has the advantages of easier implementation and lower algorithm complexity.However,the algorithm is susceptible to noise interference because of insufficient consideration of the pixel spatial neighborhood information,which makes the anti-noise performance of the algorithm poor.In this paper,the spatial information utilization rate of the fuzzy clustering algorithm was improved and the noise was shielded by using the neighborhood information to improve the noise immunity of the algorithm.The main work was as follows.(1)Non-linear mapping of pixels by using kernel function.Through the kernel function,the image pixels were mapped from the low-dimensional space to the high-dimensional space,and the original linear indivisible pixels were converted into linear divisible pixels,which improved the segmentation performance of the algorithm.(2)The original Euclidean distance was replaced by the Mahalanobis distance as a distance measure of high-dimensional space.The Mahalanobis distance could effectively describe the global relationship between two sample points,which could improve the spatial information utilization rate and the noise resistance of image segmentation algorithm.(3)The noise immunity of the fuzzy clustering algorithm was improved by using the Markov random field.The objective function of the fuzzy clustering algorithm was modified by adopting the spatial correlation and the accurate state prediction of Markov random field model.In addition,the prior probability of the Markov random field was used as the correction term of the objective function to improve the noise immunity of the fuzzy clustering algorithm.Finally,the simulation experiments were carried out in MATLAB environment.The NLFCM algorithm,LDMREFCM algorithm,FLILP algorithm,GKWFLICM algorithm and CRF-FC algorithm were compared with the proposed algorithm.The images of the Berkeley image segmentation database and the simulated noise images were segmented by using the algorithms above.The experiment used Bezdek partition coefficient,Xie-Beni coefficient,iteration number and running time to analyze objectively.The results show that the proposed algorithm can make full use of the spatial neighborhood information of image pixels,improve the efficiency of the fuzzy clustering algorithm,and effectively improve the noise immunity of the algorithm.