Using gradient tensors of the second and third orders for segmentation of images containing textures with structural redundancy

Abstract

This papers describes the effectiveness of textural features on the basis of gradient tensors in the problem of segmentation of images containing textures with structural redundancy. It is shown that the use of the bank of Gaussian bandpass filters in the formation of a feature space which is invariant to shifts and rotations of textures improves the performance of the segmentation algorithm. Examples demonstrating the application of textural features to the problem of segmentation with an unknown number of texture classes contained in the image are presented.