The Study Of Fuzzy C-means Algorithm Incorporating Spatial Information For Brain MR Image Segmentation

Image segmentation is very important to image preprocessing and patternrecognition, such as feature quantification, image registration3D reconstruction and etc.In this thesis, the brain magnetic resonance (MR) image segmentation is researched, andthe research on this paper focuses on the problem of classing the normal brain tissues.An ideal MR image is assumed to be piecewise constant. Unfortunately, theproperty is destroyed by electron and structured noises, intensive inhomogeneity andpartial volume effect (PVE). Because of the individual differences in the tissue anatomyand the slow calculating speed and inaccuracy, the current segmentation algorithms failto satisfy the need of clinical practice. Fuzzy C-means algorithm has been widely usedin brain MR image segmentation. However, the conventional FCM algorithm is used forimage segmentation, no spatial information is taken into account, leading to the FCMalgorithm get the unexpected results of segmentation when dealing with the imagescorrupted by noises. In the thesis, four improved algorithms based on spatialinformation are proposed, and the performance of these algorithms is remarkablysuperior to the conventional ones in terms of accuracy and robustness.At first, the thesis provides an overview of image segmentation and the basicinformation of MRI. Then, we present four classic algorithms respond to theshortcomings of the conventional FCM algorithm after analyzing the algorithm. Afterthat, we introduce two algorithms based on the smoothed membership, and one regardsthe sum of the neighborhoods membership as spatial information; the other introduces acontrol parameter based on the algorithm before, the basic idea is that if the pixel isnoise or edge spot, then we take no account of it. The paper proposes an improvedalgorithm based on the above two ones. This algorithm improves the control parameterwhich is determined by the distance of the neighborhood and the processing pixel.Finally, a generalized spatial Fuzzy C-means algorithm is given, that utilizes both pixelsattributes and the local information which is weighted correspondingly to neighborelements based on distance attributes. Compared with the conventional FCM algorithmand FCM-Ahmed algorithm, the performance of the GSFCM algorithm is better.