New Clustering Algorithm And Its Application In Image Segmentation

This paper covers three parts: firstly, a new visual theoretic clustering algorithmis studied to make clustering for nonlinearly separable datasets and realize imagesegmentations effectively;secondly, a new consistent possibilistic clustering with itsimage segmenting approach is studied, which can perform rational clustering andefficient image segmentations;finally, according to the above study, an imagesegmenting approach based on the clustering theory of Bayesian Ying-Yang Machineis studied to attain reasonable effect.(1) A new visual theoretic clustering algorithm is studied in this paper with itsnew cost function, which integrates visual theory together with the famous Weberlaw in biophysics to realize effective and nonparametric clustering. The experimentalresults demonstrate that this new clustering algorithm is not only effective fornonlinearly separable datasets that in general the conventional clustering approachessuch as FCM can not well cope with, but also successful in image segmentations.(2) Rooted at the exponential possibilistic model and consistent functions, aconsistent possibilistic clustering CPC with its algorithm is studied accordingly toensure choosing the rational number of clusters in a dataset and realizing clusteringwell simultaneously. Our experimental results show that CPC algorithm can makeclustering for artificial convex datasets and make image segmentations. Furthermore,in order to reduce time of image segmentations, a new framework for imageprocessing is studied, i.e., CPC algorithm is integrated together with the proposedbiased sampling theory based on Epanechnikov kernel functions to make imagesegmentations. Experimental results demonstrate that the new method can heightenthe efficiency to a certain extent without the obvious influence on the imagesegmentations.(3) In addition, in terms of the new framework for image processing, theclustering theory of Bayesian Ying-Yang Machine is combined with the proposedbiased sampling theory based on Epanechnikov kernel functions to realize imagesegmentations. Our experimental results demonstrate that the new approach can alsosucceed in making efficient image segmentations.