| Texture is an important attribute in the image, which provides substantial information for the recognition and interpretation of this image. Texture image segmentation is one of the basic research problems of the image texture. Texture image segmentation is of great significance not only to many researches of computer vision and image process but also to practical application. At present texture image segmentation comes down to all kinds of images. There are a lot of applications which appears in all fields of image process.This thesis has carried on the analysis to the basic content of texture segmentation, with its main work including conducing the research and the summary to the existing texture segmentation methods, making the improvement to these methods, and selecting a more appropriate method to apply in the heart left ventricle magnetic resonance image, obtaining quite satisfactory result. The article mainly has the following content:Firstly, A simple summary to the commonly used texture feature extraction methods as well as texture feature classification is given, and the Gabor filter is used to extract texture feature, methods of texture transformation and KPCA are also introduced into texture segmentation experiment. Experimental results show that the algorithm is better than classical algorithms.Secondly, this thesis proposes a new approach for texture segmentation based on multichannel structural feature extraction. This thesis introduces active contour model, such as Snake model and the typical geometry contour model, the principle, method of numerical implementation of which are described. And then we use the trace-based nonlinear structure tensor to extract multichannel structural feature precisely. In view of the multichannel images which are extracted, we change the edge indicator function in the level set thoery, and finally make use of a level set evolution without re-initialization to segment the multichannel images. Experimental results show that it increases the degree of segmentation accurancy and accelerate the computation speed.Thirdly, we choose appropriate texture segmentation methods and apply them to the segmentation of inner and outer contour in left ventricle tagged magnetic resonance images. This thesis proposes a new approach to segment of inner and outer contour in left ventricle tagged magnetic resonance images and LM filter banks, support vector machine as well as improned active contour model are combined in the method. Firstly we get the texture feature of the image by using the LM filter banks , and then we use the output of support vector machine (SVM) classifier relying on LM filter banks to construct a new region-based image energy term. An active contour model based on representing the contour function parametric by B-spline curves which combines the outer energy of the Chan-Vese model and the inner energy of the Snake model incorporates the new region-based image energy term to segment of inner and outer contour in left ventricle tagged magnetic resonance images. Using texture information of the images, the algorithm combines surpevised method and active contour model, while experimental results show that the algorithm gives superior performance compared with unsurpevised ones.In the last part of this dissertation, we summarized our work and analyzed the improvements to be done in the future. |
