The remote-sensing image segmentation using textons in the Normalized Cuts framework

A novel remote-sensing image segmentation method is presented in the framework of Normalized Cuts to solve the perceptual grouping problem by means of graph partitioning. In this method, texton is applied to obtain color features and texture features of remote-sensing image. Clustering of the original color values and the filter responses of the images is performed to find texton. The filter bank used in this work is consisted of an oriented edge filter at 6 orientations and 3 scales, a bar filter at the same set of orientations and scales, an isotropic Gaussian filter and a Laplacian of Gaussian filter. The histogram of texton in the small window around each pixel is used as a texture descriptor. The local similarity between these texture descriptors is measured on the texton histograms. Normalized Cut is used as a framework to solve the optimal segmentation problem with the texton feature. Efficiency and accuracy of the method are demonstrated by the texture images segmentation and remote sensing images segmentation.