Research On Fuzzy-clustering-based Method Of MR Image Segmentation

Image segmentation is the key and difficult problem of image processing and it still has no universal and effective image segmentation algorithm meeting different needs, which is also the research value of image segmentation algorithm. Fuzzy-C-Means (FCM) algorithm is a popular image segmentation algorithm among medical image processing field, but it also has many shortages. Especially when the data sample becomes big, the speed of separation becomes slow, and for medical images with much noise, the segmentation result is unsatisfactory. Magnetic Resonance Imaging (MRI) plays very important role in clinical diagnosis and MR image segmentation is a very popular topic for medical information processing and artificial intelligence. This thesis chooses brain MR images as the research target, uses the C-Means algorithm for image segmentation, and puts forward an improved algorithm to optimize the performance of the system.For traditional Fuzzy-C-Means Clustering (FCM) algorithm in image segmentation (especially medical image segmentation) existing the problems of large amount of calculation and running time longer and the unfavorable situation of samples leading to unfavorable clustering results, it proposes an improved method. It uses clustering center obtained by fast speed of the K-Means convergence as the initial clustering center of FCM algorithm, reduces iterative times needed by FCM algorithm convergence to improve the speed of image segmentation. With the constraint condition that the sum of some samples of FCM for kinds of clustering membership grade is one, it changes the sample number sum of all samples for kinds of clustering membership grade to improve the accuracy of brain magnetic resonance image segmentation.To improve the segmentation speed of FCM algorithm and anti-noise ability of image noise, it adopts the corresponding improvement on rapid FCM.The experiments show that the improved FCM algorithm in improving the segmentation rate is also good for noise image denoising effect.Because traditional clustering algorithm is adapted to specific structure of data, in order to meet the arbitrary shape cluster analysis of the data requirements, this paper in-depth study of nuclear methods, based on its introduction of FCM clustering to construct a fuzzy Nuclear C-Means (Kernel Fuzzy C-Means, KFCM) image segmentation algorithm. And slow arithmetic operations, as well as the shortcomings of poor anti-noise is proposed to improve fuzzy kernel clustering algorithm for image segmentation. In order to eliminate image noise , the algorithm combine image wavelet denoising and median filtering approach at the same time. In order to improve the computing speed and enhance its robustness, K means algorithmused as initial clustering center for Kernel Fuzzy C-Means , and return a constraint method to its optimization. By a large number of data partitioning experiments, it results from the segmentation accuracy and run time to prove the validity and accuracy of this method.