Brain Image Segmentation Based On The Improved FCM Algorithm And MRF

Segmentation of brain magnetic resonance(MRI)images plays an important role in computer-aided diagnosis and clinical research.However,segmentation of brain images is a challenging task due to problems such as noise and uncertainty in the boundaries between different tissues in the brain image.Therefore,the study of brain MRI image segmentation has certain practical significance.In the image segmentation algorithm,FCM(Fuzzy C-means Clustering)is a classic algorithm suitable for brain image segmentation.The essence of the FCM algorithm is to maximize the similarity between objects belongs to the same cluster,and minimize the similarity between the different clusters,and simultaneously calculate the minimum value of the objective function,thereby achieving fuzzy division of each pixel.FCM algorithm and its improvement are studied in this paper,find the following shortcoming:Firstly,the segmentation process only considers the situation of pixels in self circumstances,which makes the segmentation result very sensitive to noise;the second is that the relationship between the neighboring pixels and the central pixel is not fully considered,that is,the influence of the difference between the pixels is not utilized,resulting in The inaccuracy of the segmentation result.In order to overcome the above shortcomings of the FCM algorithm,the weight of the neighborhood pixel to the central pixel is introduced to improve the FCM algorithm.First,the clustering pixels are weighted,and the gray value of the pixels to be clustered is calculated by combining the gray distribution and similarity of neighborhood pixels.And then on this basis,using the formula of the coefficient of heterogeneity proposed in this paper to calculate the influence degree of the neighborhood pixel on the central pixel,and then the distance between the neighborhood pixel and the central pixel are calculated respectively,and the membership degree of the control center pixel is obtained,thus the objective function could be controlled.In this paper,Markov random field is introduced to suppress noise,and a new membership function is obtained by combining the local prior probability of Markov random field with membership degree,so as to improve the segmentation effect.Finally,this algorithm is implemented using MATLAB and compared with the existing FCM algorithms.By comparing the experimental results of various methods with relevant evaluation criteria,it is proved that the proposed algorithm can achieve the segmentation of brain images with accurate segmentation results.