Research On Density Clustering Algo Rithm And Its Application In Image Segmentation

Image segmentation technology is the basic work of image processing and analysis.Traditional image segmentation technology mostly uses pixels as the basic unit.Superpixel method is a technology that forms a series of pixel regions from adjacent pixels with similar features such as color,texture and brightness.Image segmentation based on superpixel not only retains the important content of image segmentation,but also does not destroy the boundary information of the target object,which can greatly simplify the image representation and reduce the complexity of subsequent image processing tasks.The key point of image segmentation based on superpixel is superpixel generation,and clustering is an effective method to generate superpixels.Therefore,this thesis focuses on density peak clustering(DPC)and Neutral mean clustering and their applications in superpixel generation and image segmentation.The work content of this thesis mainly includes the following three aspects:(1)DPC is a clustering method based on local density.The two basic factors that influence the effect of DPC are local density definition and class center selection.Aiming at the problem that the classical DPC does not consider the distribution of the sample points in the neighborhood when defining the local density,and it is unable to automatically select the class center,an automatic threshold density peak clustering algorithm based on distribution is proposed,and it is applied to the superpixel generation.Experiments show that the new algorithm can not only automatically select clustering centers,but also obtain higher classification accuracy and higher quality of super-pixel images compared with the existing algorithm.(2)Considering that the initial parameters of NFC segmentation method need manual intervention,and the effect of spatial information on noise and segmentation effect is not considered,a neutral fuzzy clustering image segmentation algorithm(DP-NLNFC)based on non-local information is proposed.This method uses the density peak algorithm to initialize the parameters reasonably.Experiments show that the new method has better noise resistance and more accurate segmentation.(3)By combining the advantages of these two algorithms,a hybrid image segmentation method based on density clustering is proposed.This method firstly preprocesses the image by using the automatic threshold density peak clustering based on distribution to obtain the super-pixel image,and then selects the reasonable initial parameters based on the density peak,and then clustering the image based on the super-pixel image by using DP-NLNFC to obtain the image segmentation results.Experiments show that the proposed method can achieve high quality and high efficiency in image segmentation.