a review study on automated content-based image retrieval

in this research, a new segmentation alrorithm was proposed and demonstrated using texture images. for the segmentation algorithm, complex wavelet transform and kernel principal component analysis was used for feature extraction and dimensionality reduction. kernel principal component analysis using gaussian radial basis function kernel was shown to be elective when capturing localized feaures corresponding to clusters. the kernel mean shif clustering procedure was introduced for cluster analysis. the gauussian radial basis function mean shift kernel was shown to be adaptive to the shape of anisotropic clusters and it would, in theory, lead to a better representation of clusters as compared to its linear counterpart. although the kernel mean shift clustering procedure has produced only a slight improvement in the experiment results, it is belived that with the desired properties, it could be applied to other data analysis problems.