A Study of Weighted Block-based Fuzzy C-means Clustering and Co-correlate Histogram Technique for Human MRI Image Segmentation

• Main Authors: His-Yun Ho, 何錫昀
• Other Authors: 吳憲珠
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/7x8h5e

碩士 === 國立臺中科技大學 === 資訊工程系碩士班 === 101 === The Magnetic Resonance Images (MRI) is one of the common ways to display the cerebral structures. The Parkinson''s disease may lead to the cerebral cells atrophy that includes Caudate nucleus, Putamen and Thalamus…etc. Segmentation result of the cerebral cells can more effectively and accurately help doctors to diagnose diseases and shorten time of diagnosing. Therefore, this thesis proposes two automatic schemes processing the automatic MRI image segmentation of cerebral cells in applying the research of medical image. The first proposed scheme in this thesis presents MRI image segmentation based on Fuzzy C-means clustering (FCM) algorithm with automatic parameter adjustment. The proposed method performs block-based clustering by modifying the FCM algorithm. It determines the number of clusters automatically by applying the MRI images information in the initial step. Additionally, the proposed method uses the weighted vectors corresponding to each cluster of the MRI image. Firstly, the proposed method would automatically determine the initialized cluster-center members according to the gray values distribution of MRI image. Simultaneously, it would automatically generate the weighted vectors from Gaussian function for the membership matrix of the clusters in each iteration. Finally, the results of clustering segmentation are performed by the morphology and threshold operators to extract the main cerebral tissues. The implementation of proposed method does not need high computational time to iterations and according to the MRI image’s gray level distribution; it could produce initialized cluster-center members. The second proposed scheme in this thesis presents an application of co-correlate histogram to MRI images segmentation and contrast enhancement techniques. First, establish co-correlate histogram which is the intensity distribution of pixel values in spatial information of the MRI image and then analyze quadrants of each layer within the co-correlate histogram. Use the co-correlate histogram to automatically enhance the MRI image; it’s according to quadrants of each layer of analyzing the co-correlate histogram to enhance the pixels. Then, establish co-correlate histogram of enhanced MRI image and apply its analytic quadrants of each layer to automatically segment the cerebral cells in MRI image. From the experimental results, the proposed method can automatically contrast enhancement with the original one and segment the cerebral cells which are in the similar gray value region.