Supervised texture classification of remotely sensed imagery using rotation invariant moments in spatial frequency domain

摘要

[ABSTRACT] A method for representation of textures within a small part of image is proposed. This method consists of the following procedures: 1) histogram stretching based on statistical characteristics of the target area; 2) windowing with Hanning function for reduction of edge effect on the FFT; 3) calculation of power spectrum using the 2D-FFT; 4) quantification of the spatial spectrum pattern by using various orders and repetitions of Zernike moments; and 5) normalization by Euclidean distance from the origin in the Zernike moment space. Consequently, the texture of the target area can be uniquely represented as a vector from the origin to a point on the surface of the unit hypersphere in the Zernike moment space. The method has invariant property for image translation and rotation. In this article, we describe the details of the method, and also demonstrate that the method can make supervised texture classification effective more than conventional classification method.