The Application Of Local Binary Fitting Level Set Method For Medical Image Segmentation

Image segmentation, which is a basic part of image processing, is the premise of image understanding and target recognition, and has been a hot and difficult problem in the field of image processing and computer vision. Recently, many researchers have done a great deal of efforts to improve the performance of active contour methods which implemented by level set. Classical methods like PC、LBF、LIF be proposed, but we can’t obtain a satisfactory results if we use these algorithms to deal with intensity inhomogeneity images or weak edges. Methods like PC、LBF、LIF are two phase model. Real images usually contain multiple homogeneous target area which can not be segmented by the two phase model. So, multiphase models have important research significance. The main works can be summarized as follows:(1) In this thesis a new LBF model based on the local gradient have been proposed. By incorporating the LBF model with the gradient in the neighborhood of a contour, we can have a more accurate approximation to the original image. Experiments on intensity inhomogeneity images or weak edges demonstrate the advantages of the proposed model over the old one in terms of accuracy. In the end, we do a combination of the improved model with the multiphase model which use N level set functions express2N phase. In order to have a comparison with the old LBF model’s multiphase extention, several experiments have been done. Experiments demonstrate the improved model with the multiphase model which use N level set functions express2N phase have better segmentation results.(2) In this thesis a new multiregion competition algorithm which control the level set functions’define order be proposed. The new multiregion competition algorithm can segment images which contain totally over lap target areas under whatever initialization. One algorithm judging the need of exchange the level set functions’order been added after each iteration. We exchange the level set functions’define order if the later defined level set function’s positive region completely surrounded by the earlier defined level set function’s positive region. A detailed introduction to the combination of the new multiregion competition algorithm and the new LBF model been done. Experiments show that we can get better segmentation results if we exchange the level set functions when the later defined level set function’s positive region completly surrounded by the earlier defined level set function’s positive region.(3) The old multilayer level set segmentation do not have a clear approach to deal with the parameters li and m, but li, and m are importment to the algorithm’s realization. So in this thesis, a preprocessing step have be added, and the value of li, and m can be decided. Then, we combine LBF with the new multilayer level set segmentation, a comparison with N level set functions express2N phase multiphase model and multiregion competition algorithm be done respectively. At last, we combine the new LBF model based on the local gradient with the new multilayer level set segmentation, a comparison with N level set functions express2N phase multiphase model and multiregion competition algorithm be done respectively.