Somayeh Ahmadi - Semnan University

In this research, a new segmentation algorithm wasproposed and demonstrated using texture images. Forthe segmentation algorithm, complex wavelettransform and kernel Principal Component analysiswas used for feature extraction and dimensionalityreduction. Kernel principal component analysis usingGaussian radial basis function kernel was shown To beelective when capturing localized featurescorresponding to clusters. The Kernel mean shiftclustering procedure was introduced for clusteranalysis. The Gaussian radial basis function mean shiftkernel was shown to be adaptive to the shape ofanisotropic clusters and it would, in theory, lead to abetter representation of clusters as compared to itslinear counterpart. Although the kernel mean shiftclustering procedure has produced only a slightimprovement in the experiment results, it is believedthat with the desired properties, it could be applied toother data analysis problems.

Principal Component Analysis, Kernel, Gaussian Radial Basis Function, Content-based Image Retrieval, Kolmogorov-Smirnov