Research On Brain MR Image Segmentation Based On Fuzzy Means Clustering Algorithm

Image segmentation is a technique to extract the interested object from target image in order to assist image processing and analyzing. Image segmentation has been applied to a lot of fields, such as computer vision, pattern recognition, medical image process and so on. The clustering-based methods are very important and wide-used in image segmentation, and the most commonly used method is Fuzzy C-mean (FCM) clustering, which doesn’t need setting any threshold or get people involved. For the clustering accuracy of the traditional FCM algorithm is low and susceptible to noise in image segmentation, this dissertation makes improvements on it.(1) Gaussian kernel function is introduced to the FCM algorithm in this dissertation to instead Euclidean distance of the traditional FCM algorithm, converting low dimensional nonlinear systems into high dimensional linear ones, which reduces the complexity of the problem. To avoid the problem proposed by Dave that the noise distance value δ is constant, the function of 8 is improved in this dissertation to change with the pixels.(2) The uncertainty of the membership degree makes clustering result less effective. Therefore, in this dissertation, information entropy is introduced as a method to measure the quality of clustering.(3) In order to improve the anti-noise performance of the fuzzy clustering algorithm, a novel clustering objective function is proposed from the perspective of neighbor membership constraints, and a fuzzy mean clustering image segmentation algorithm based on neighbor membership constraints is obtained. The constraint can be used to restrict or even block the membership functions with impossible or undesirable structures.Comparison experiments of the synthesis image and brain magnetic resonance image are designed in order to verify the effectiveness and practicability of the improved algorithm proposed in this dissertation. Results show that the algorithm proposed in this dissertation improves the clustering accuracy and has a good performance of noise reduction compared with the traditional clustering algorithm.