Research On Guided Filter Based Fuzzy Clustering Algorithm And Its Application In Image Segmentation

Accurate image segmentation is of great significance for image analysis.Due to the advantages of simplicity,flexibility,and efficiency,prototype-based fuzzy clustering algorithm has been widely used in image segmentation.However,the existing fuzzy clustering algorithms are difficult to get ideal image segmentation results.The classical fuzzy clustering algorithm obtains poor segmentation results for the noise image,because it only considers the intensity value of the pixel.Through fuzzy clustering algorithm with mean filter or median filter can effectively suppress noise,it will lead to the edge information loss.At present,the guided filtering based fuzzy clustering algorithm can effectively remove noise and preserve the edge information of images,but it is only used for the gray image segmentation.This thesis conducts the research on the guided filter based fuzzy clustering method.The algorithms for gray image segmentation,color image segmentation and general data clustering are designed.The work of this thesis is listed in detail as follows.For the gray image segmentation,a parameter optimization guided filter based fuzzy clustering method is first proposed in this thesis.In this method,the influence factor is defined to adjust the guidance image,and different influence factor values are set to improve the segmentation results for different levels of noise image.This adjustment on guidance image is mathematically proved to be more efficient than the direct change of the filter parameter.Then,the morphological reconstruction method is employed to reconstruct the noisy image so as to improve the segmentation results of high noise image and simplify the selection of influencing factors.In the test,the above methods achieve ideal results on the image with Gaussian noise or Rician noise.For the color image segmentation,a novel weighting multi-channel guided image filter method is first proposed in this thesis.In this method,each channel of the color guidance image is utilized to guide the filtering for the gray input image independently.The final output image is defined as the weighted sum of the filtering results of multiple channels,where the weight for a channel is defined according to the variance of the image pixels in a local window.Compared with the existing multi-channel guided filtering method,the proposed method obtains more accurate filtering results and has lower time complexity.Then,based on the weighting multi-channel guided image filter,this thesis presents a fuzzy clustering method for color image segmentation.Finally,this thesis develops the multivariate morphological reconstruction based fuzzy clustering algorithm with a weighting multi-channel guided image filter.In this algorithm,the morphological reconstruction method is employed to reconstruct the color noisy image so as to further improve the segmentation results of color noise images.The experiment results demonstrate that the proposed clustering algorithms obtain the ideal segmentation results on the color image with Gaussian noise or Salt & Pepper noise.For the general data clustering,this thesis first designs a new filter window to determine the neighbor of each data,and to guarantee that the neighborhood relationships between data samples are symmetric.Then,based on this data filter window,this thesis designs a fuzzy clustering algorithm framework with guided filtering for general data clustering.What's more,the modified versions of the fuzzy c-means clustering algorithm,the fuzzy clustering algorithm with the entropy of attribute weights,and the kernel fuzzy clustering algorithm improved by this framework are presented in this thesis.The experiment results show that the proposed framework can effectively improve the results of the above three fuzzy clustering algorithms on nonlinear datasets,UCI datasets and image datasets.