Fuzzy Clustering Based On Local Spatial Information Algorithm Research

Image segmentation is an essential research area in the age of big data information.Clustering in segmentation algorithms is of great concern because of its capability to extract significant features from large groups of information.Among them,fuzzy clustering is more in line with the division rules of actual things and can address the problem of easy oversegmentation,which is extensively used in real life.However,conventional fuzzy clustering algorithms do not exploit the neighbourhood information of the image,and use Euclidean distance measurement,so they are very sensitive to noise and have low segmentation precision and weak resilience.In response to the above problems,this paper improves the image from its local spatial information and membership,and applies multi-scale to image filtering and segmentation as a way to improve the performance of the algorithm.The details are as follows:(1)Existing algorithms have limitations for when the image is heavily contaminated with noise.This thesis proposes a robust kernel fuzzy clustering method using a combination of multi-scale filtering and fuzzy clustering.Firstly,the new target object function is built from the local spatial and weight information of the image,and a Gaussian kernel metric is used to convert the indivisible complex linear data into decomposable data in a highdimensional space.Secondly,the neighbourhood information of the membership is used to correct it twice to improve the classification precision.Finally,the image to be segmented is blurred several times with a Gaussian filter in order to construct a multi-scale space,and the final result of the clustering of the upper image is used as a guide for the clustering of the lower image according to the pyramid model,which prevents completely random initialisation and thus improves the robustness of the method.The technology provides superior segmentation on both real and simulated images,and is robust to noise.When measured against experiments with different levels of noise interference on multiple datasets.(2)It is feasible to obtain images with different details in a multi-scale space-based model,but it also creates the problem of needing to consume a lot of time.The time complexity is an important determinant of the practical worth of the algorithm.The multiscale space is therefore reformulated to find a multi-dimensional model that clusters the resulting multi-scale images in a single pass.The Gaussian kernel is retained in fuzzy clustering to capture intricate pixel associations,and the weighting factors are re-combined based on multiple local proximity relationships,and solve it quadratically using the likelihood and mean of membership to enhance the probability of attribution for each point.Different experiments were conducted with each of the remaining seven algorithms under the augmentation of the two filters to reflect the actual segmentation performance of the algorithm in terms of both visual and data,so that the algorithm in this thesis provides better resistance to common heavy noise also in terms of time.