Research Of MRI Image Segmentation Based On Support Vector Machines

Medical image segmentation is an indispensable technique for extracting quantitative information of the specific tissues, and it is also the premise of image three-dimensional (3D) reconstruction and visualization. Segmented images can be applied to numerous research fields including diagnoses of pathological tissues, and observation of anatomical structures, computerized operation, and 3D visualization, etc. As the foundation of the three-dimensional reconstruction of the head model, the denoising and segmentation for MRI image have been down in this paper.Since noise existed to some degree, pre-processing for original MRI image has to be executed in order to acquire high quality image. It is also propitious to improve the accuracy of classification for satisfactory segmentation result. In the process of denoising, Adaptive Template Filtering Method (ATFM) is instructively researched on the bases of the traditional algorithm. Experiment indicated this method restrain the noise effectively.More and more new methods have been introduced in the field of image segmentation recently. Machine learning has been a topic area now. Traditional machine learning can not control complexity of model since it followed the principle of Empirical Risk Minimization (ERM). Therefore, it can come forth over-fitting and under-fitting problem resulting in low generalization ability, especially for the case of finite samples. Unfortunately, the number of training samples is actually finite and the data dimension is high, thus the performance of traditional pattern classification algorithms can not meet the need of medical image segmentation. Taken into account the factors of the good generalization ability of support vector machine in small samples, nonlinearity and high dimension space and the features of medical images, this dissertation deeply studies support vector machine methods and their application in head MRI image segmentation, and successfully obtains the edge of the multi-objects and volume data. This part of work makes the foundation for 3D head model reconstruction.For MRI image, it is not enough for multi-object segmentation just depended on gray degree. So the feature extracting method of combined textures and gray features are researched. In order to realize dimension reduction and eliminate redundancy information, Principal Component Analysis (PCA) is used for image feature extraction. Theory and experiment results indicate that the segmentation performance by combining PCA with SVM is more effective than using SVM only.