Research On Clustering Algorithm And Its Application In Texture Image Segmentation

There are many division problems in natural science and social science.With the development of science and technology,people ' s requirements for division are also rising.It is difficult to divide accurately only by experience and professional knowledge,and the main research branch of division problem is clustering algorithm.It classifies the similar samples into one category and divides the elements with large differences into different categories.This paper studies a variety of classical clustering algorithms,including Density-Based Spatial Clustering of Applications with Noise(DBSCAN)and Affinity Propagation Clustering(AP).Since the clustering algorithm is simple and efficient,it is widely used in image segmentation.In this paper,the improved AP algorithm is applied to texture image segmentation,and good segmentation results are obtained.The specific research contents are as follows:Firstly,in view of the strong dependence of DBSCAN clustering algorithm on neighborhood radius parameter values,a topological similarity DBSCAN clustering algorithm(TS-DBSCAN)is proposed.Based on DBSCAN algorithm,TS-DBSCAN algorithm establishes two functions: Cluster function and Combine function.The operation of Cluster function includes three steps.Firstly,the centers are obtained by the weight of average distance.Secondly,the parameter values corresponding to each cluster center are obtained by topological similarity.Finally,the obtained centers and parameter values are used to obtain multiple small clusters.The Combine function combines small clusters by density reachability and density connection to obtain the final clustering results.Simulation results show that compared with other clustering algorithms,the average normalization and accuracy of TS-DBSCAN algorithm are increased by 6 % and 8 %,respectively.Secondly,aiming at the shortcomings of the AP algorithm that it cannot identify non-convex data and has strong dependence on the bias parameter Preference value,an adaptive nearest neighbor propagation clustering algorithm(GA-AP)based on universal gravitation is proposed.Based on the traditional AP algorithm,this algorithm uses the universal gravitation to calculate the similarity(gravitation)between data,and uses the information entropy and adaptive enhancement algorithm to obtain the weights of the correct cluster sampling points and the wrong cluster sampling points in each cluster,so as to reduce the dependence of the algorithm on the Preference and the number of wrong cluster sampling points in the cluster.The simulation results show that compared with other clustering algorithms,the average normalization and accuracy of GA-AP algorithm are increased by 6 %.Thirdly,GA-AP algorithm is applied to texture image segmentation.The implementation process includes three stages: image preprocessing,feature extraction and clustering.The comparative experiments by using artificial images and Brodatz texture images show that the GA-AP algorithm has higher segmentation accuracy than other traditional algorithms.