Research Of Image Segmentation Based On Fuzzy Clustering And Level Set Methods

Image segmentation is a very important part of the Image processing. in the process of the analysis and processing for image, the segmentation is its essence, determines the final image quality analysis and image understanding. While people have done a lot of segmentation research, but there is no general theory of the division. Proposed segmentation algorithm has been mostly issue-specific, so review of image segmentation methods and evaluation criteria is different, but for each type of image segmentation, we choose the best according to the method of application requirements, while still keep explore new segmentation and segmentation image segmentation theory to meet the new demands made by, and this is where the purpose of this thesis topic.Fuzzy C-means algorithm is the most perfect and the most widely used algorithm which is based on the objective function of the clustering algorithm theory, The algorithm introduced the concept of fuzzy membership degree to the image pixels, to retain more of the original image information. Fuzzy C-Means clustering method also has the problem of setting the threshold to avoid the merits of improving the scope of the algorithm. Fuzzy C-Means clustering algorithm has good convergence, while more than one branch can be used to solve division problems. FCM algorithm is based on fuzzy set theory proposed, therefore, of randomness and fuzzy image of the division has a good effect. However, during the image segmentation algorithm is needed to determine in advance the number of categories, sensitive to noise, easy to fall into local minimum.Level set method, greatly promoted the development of active contour models, level set method combined with the curve evolution model to overcome the many inherent shortcomings of traditional models, which greatly expanded the scope of application of active contour models. But the geometric active contour model evolution curve and lack of self-adaptive boundary leakage problem will remain, especially for soft-edge medical imaging, segmentation results are poor.In view of these considerations, the paper introduced by the traditional spatial context information fuzzy C-means algorithm, to improve anti-noise performance of the algorithm; using particle swarm algorithm to the traditional fuzzy C-means algorithm to improve; with geometric active contour model to maintain the level set in the distance inadequacy of their model to improve it. Main tasks are:Proposed kernel-based and space domain information neighboring fuzzy C-means image segmentation algorithm. Between adjacent pixels in the image there is a strong correlation, but the traditional FCM algorithm during the image segmentation considered only grayscale or color images color characteristics, while ignoring the wealth inherent in the image spatial information, making it more sensitive to noise, but also making the final out of the area often do not split a row, through the introduction of the membership function to solve; with the kernel-induced distance instead of the traditional FCM algorithm in the Euclidean distance to a given linear space from high-dimensional space into linear distance. Thus, in the original low-dimensional space of complex linear nonlinear problems can be treated. Therefore, this article will be the neighborhood kernel function and spatial information into fuzzy C-means algorithm. By fuzzy C-means algorithm on the original objective function of the membership function of the distance metric correction and improved algorithms to improve the existing anti-noise performance. Algorithm by modifying the value of membership, to improve the convergence speed.FCM algorithm is essentially a gradient descent based local search algorithm, there is a greater dependence on the initial value, such as inappropriate choice of initial value, the algorithm will converge to a local minimum but not global optimal solution. In this regard, this paper particle swarm global optimization features will be applied to the traditional fuzzy C-means to determine the initial value, we propose a new and improved particle swarm optimization based on fuzzy C-means clustering algorithm for image segmentation , the algorithm first with a fast convergence of the hard clustering algorithm for image clustering has been hard clustering center, then as a reference value, initialize the PSO algorithm for global search, while the introduction of the PSO algorithm to repulsion between particles control the population diversity to avoid premature convergence problem. Finally, the global optimization results as a follow-up the initial cluster centers FCM algorithm, the image clustering segmentation. And image segmentation based on the traditional FCM algorithm, the algorithm less iterations, convergence speed, better image segmentation.Proposed to maintain the distance based on improved adaptive level set image segmentation model. First introduced the level set curve evolution theory and level set model for image-based edge strength and region-based image information of the geometric active contour model is discussed in detail in the two models leads to keep the distance on the basis of the level set image segmentation model, and the model improvements. First of all, keep the distance the level set model is superior in many respects, although the boundary-based and region-based active contour models, but there are still evolution curve of the initial position of the boundary lead to dependence and weak boundary leakage problem. In this paper, keep the distance to improve the model, the introduction of region-based symbolic pressure function, to some extent out of dependence on the initial position, the initial curve in the image of any place, can automatically detect the image boundaries. Second, to maintain the level set model from the introduction of region-based symbolic pressure function, although to a certain extent from the initial position of dependence, but when the evolution of the boundary, there will still be the boundary leakage problem, so on this basis, the introduction of a new image information associated with variable weights to the image information according to the direction of motion adaptive changes, while improving the detection of weak boundary layers and the ability to profile, but also accelerated the evolution speed.Papers on image segmentation and fuzzy clustering segmentation algorithm geometric active contour models have been studied and discussed, and through various means to improve the original algorithm, images of different paper presents experimental results show that the effectiveness of the algorithm.