Font Size: a A A

The Clustering Algorithm And Its Application In Image Segmentation Based On Gaussian Mixture Model

Posted on:2014-01-09Degree:MasterType:Thesis
Country:ChinaCandidate:X WangFull Text:PDF
GTID:2248330395492116Subject:Applied Mathematics
Abstract/Summary:PDF Full Text Request
Image segmentation is a key technology of digital image processing, andthe Gaussian mixture model is commonly used in image segmentation model,which played an important part in image segmentation. The application ofclustering algorithm in the areas of image process technology also promote thedevelopment of clustering algorithm itself, and makes it an important hotspot inscientific research areas.In this paper firstly we analyzed the EM Algorithm of estimating parameterin Gaussian mixture model and we found a faultiness in EM Algorithm whichis that EM Algorithm is an local algorithm, so it is very often to reach a localoptimum solution which makes the choosing of initial value really important tothe outcome result. In order to reach a better clustering result we need toinitialize the EM Algorithm effectively. In this paper we first give a rough blocksort of mixture data through k-means clustering and then we estimate the initial value which is as an initial value of EM Algorithm based on the block sorts. Thenumerical simulation experiment shows that this method is much better than theoriginal EM Algorithm.Secondly we researched the model selection problem which is a problemabout how to select the number of Gaussians in Gaussian mixture model andanalyzed the Bayesian Yin and Yang harmony learning system algorithm (BYY).The advantage of dynamic regular BYY learning algorithm with back directstructure is that as long as there are more number of component design ofGaussians mixture model than real Gaussians, then the unnecessary Gaussianswill be compressed to zero during the learning process, and so complete themodel selection. Numerical simulation experiment proved its effectiveness.In the last, in this paper we put the self-adaption dynamic regular BYYalgorithm with back direct structure applied to the image segmentation and theresult shows that it can find the right number of Gaussians clustering but thesegmentation is not ideal enough. Then Using dynamic regularization BYY withbackward structure learning algorithm on the number of categories, we will getbetter segmentation effectiveness by doing again image segmentation combined with EM algorithm.
Keywords/Search Tags:Gaussian mixture model, clustering analysis, image segmentation, EM algorithm, Bayesian Yin and Yang harmony learning system algorithm
PDF Full Text Request
Related items