Texture extraction of high resolution remote sensing image based on the characteristic of image wavelet coefficients

This paper presents a method for feature extraction of high resolution remote sensing image which is based on the statistical model of the marginal distribution of wavelet coefficients. First, the wavelet is used to transform high resolution remote sensing images into frequent domain. Then, Generalized Gaussian density(GGD) is used to accurately model the marginal distribution of wavelet subband coefficients (wavelet coefficients histogram) followed by the establishment of the remote sensing image texture feature vector. Finally, the Kullback-Leibler distance (KLD) is computed between the texture feature vectors as similarity measurement(SM), and the output is ordered by the result of the SM. Experimental results show that this method is effective and efficient, and the image feature can be well represented by this texture feature vector. The advantage of this method is that the SM step can be computed entirely on the estimated model parameters, which has solid theoretic background, so that it can meet the requirements of the CBIR application.