Image Segmentation Method Based On Superpixel And Spectral Clustering And Its Application

Image segmentation is to extract the target area in the image,which has an important impact on many high-level image processing,such as target recognition,object tracking,and scene analysis.Because the image can be easily converted into the structure of the graph,the map can describe the spatial position information of the pixel well.Therefore,the image segmentation method based on spectral clustering is an effective image segmentation method currently.In order to reduce the number of nodes in the graph,this paper introduces the concept of superpixels,and proposes some improvement methods for the deficiencies of spectral clustering algorithms.Finally,image segmentation is applied to remote sensing image processing,which reflects a certain application value.The main research contents of the paper are as follows:First,the principle and implementation of spectral clustering algorithm are stud-ied.Experiments verifly the influence of the selection of parameters ? and feature vectors on the clustering effect,and the advantages of spetral clustering algorithms over oth-er conventional clustering algorithms such as k-means and DBSCAN(Density-Based Spatial Clustering of Applications with Noise).Secondly,the method of selecting the feature vector by using the maximum vari-ance between classes and using CFSFDP(Clustering by Fast Search and Find of Density Peaks)algorithm to determine the numbers of clusters is proposed,which improves the segmentation effect of the region with no obvious difference in the image,and makes up for the artificial input cluster numbers in the spectral clustering algorithm.After that,the principle and implementation steps of the SLIC(Simple Linear Iterative Clus-tering)superpixel generation method are analyzed.The influence of the parameters K and m on the superpixel generation result is verified through experiments,and on this basis,an image segmentation method based on SLIC and improved spectral clustering is realized.Finally,this paper applies image segmentation to ice and snow type identification in the Antarctic ice sheet.In the implementation process,first of all,using the coastline data to remove the sea part.Then,according to the backscattering coefficients and texture features,the image segmentation method is used to segment the image,and the maximum likelihood estimation method is used to label the segmented regions into dry snow,infiltration zone,wet snow,and shadow.Finally,based on the area and the neighboring relationship,regional consolidation is performed to obtain the final result.Experiments show that this method can accurately achieve the separation of wet snow,dry snow and infiltration zone.