Density Peak Clustering Algorithm Based On Neighborhood Mutual Information

Cluster analysis is an important research direction in data mining.Cluster analysis divides datasets into clusters based on the similarity between data,allowing some data with high similarity to be classified into one class and those with low similarity to be classified into different clusters.In today’s era of data explosion,because clustering analysis can reveal the inherent,potential and unknown knowledge,principles or rules in the real world,and reveal the laws and patterns of data,it is widely used in data mining,pattern recognition,machine learning,information retrieval,image analysis,computer graphics and other fields.Now,it has proposed different clustering strategies,such as zoning,layering,distribution based Based on several densities.Density based clustering(DPC)is a clustering algorithm published by Rodriguez and Lario2014 in Science.This algorithm determines the density center through a decision map of the density peak points and center offset distance of the cluster,and then efficiently allocates sample points through the density center.However,the DPC algorithm still has the following limitations:(1)All sample points lack precise partitioning,and although it is efficient,there is no clear threshold for the intermediate points in the decision graph,resulting in insufficient clustering accuracy.(2)Due to its high time complexity,it is not suitable for clustering analysis with large datasets.Considering the accuracy and time complexity of the algorithm,this paper proposes an improved density peak clustering algorithm(NMIDPC)based on neighborhood mutual information.Firstly,a density measurement method based on neighborhood mutual information is proposed,the local density is recalculated,and the points in the neighborhood of the density center are subdivided according to the neighborhood density,which improves the decision-making accuracy.Secondly,a density peak clustering algorithm based on neighborhood mutual information is proposed,which can find the division of each point by continuously zooming in and out of the neighborhood of the density center to improve the clustering accuracy.Finally,for high time complexity problems,the boundary domain acceleration algorithm is used to accelerate the calculation of points located on the boundary domain in the decision graph,reducing time complexity.The experimental results on the UCI dataset show that in most datasets,the NMIDPC algorithm not only improves accuracy but also reduces time complexity.On this basis,this algorithm is applied to image segmentation,and a density peak clustering image segmentation algorithm based on neighborhood mutual information is proposed.First,we will use neighborhood mutual information to merge a small segment into another compact data to obtain a better segmentation effect.Then,the image is segmented by matching the fields and using the segmented variance formula.This method can use variance to calculate the distance between the selected segment and other ends,allowing us to find matching segments.Finally,we propose a density peak clustering image segmentation algorithm based on neighborhood mutual information.This algorithm can complete image processing without a large number of iterations,reducing the time complexity of the algorithm and improving the accuracy of image segmentation.Experiments show that the density peak clustering algorithm based on neighborhood mutual information performs better than the image segmentation algorithm based on classical clustering in the image segmentation dataset.It provides support for the application of density peak clustering image segmentation algorithm based on neighborhood mutual information in actual scenes.