Marginal distribution covariance model in the multiple wavelet domain for texture representation

Abstract

Filters are the commonly used techniques for texture feature extraction. In this paper, we present a texture feature extraction method based on multiple wavelet filters. For modeling the multiple wavelet coefficients, we develop Marginal Distribution Covariance Model (MDCM) in which the data points are projected into Cumulative Distribution Function (CDF) space and then the covariance model is constructed in the CDF space. MDCM can capture the dependence of variables, so it can be applied to model the texture features, in which the dependence exists, such as image intensities, color features and wavelet filter features. According to the characteristics of different wavelet filter features, we construct the different MDCMs in the three wavelet transform domains: Orthogonal Wavelet Transform (OWT), Dual Tree Complex Wavelet Transform (DTCWT) and Gabor Wavelet Transform (GWT). Experiments show the proposed method which uses multiple wavelet features can provide a promising performance compared with the state-of-the-art methods.