Clustering of Texture Images Based on Fourier Power Spectrum and Minute Shape Features

Abstract

In this study, an approach for clustering various texture images has been derived using both Fourier power spectrum patterns and a shape-pass filter bank, and the experiments have been carried out. Clustering of texture images have been tried so far, but it was necessary to restrict texture images to some class to get good results. Furthermore, both structural and statistical viewpoints are essential to the fundamental approaches for texture images, but a new viewpoint is needed to fold the human sensitivity to texture images. In this approache, the results coinciding with human sensitivity to some degree have been got by clustering using Fourier power spectrum patterns of circle, doughnut, ellipse, twin, cross, line and other periodical patterns, then the shape feature vector of nine shape elements, black/white line, black/white pepper, black/white roof, black/white snake, and cliff.