An Improved Minimum-Distance Texture Estimator for Speckled Data Under the $$\mathscr {G}^0$$ Model

Abstract

The G 0 distribution is an apt model for speckled data, such as SAR imagery, because it possesses the ability to characterize areas with different degrees of texture. In the monopolarized case, this distribution depends on three parameters: the texture, the scale, and the number of looks. The first one is related to the roughness of the image, so its estimation deserves special attention. This paper proposes and compares estimation methods of the texture parameter in intensity format. We treat the multilook case. The proposal is to estimate this parameter by minimizing the triangular distance between the G 0 density and an estimate of the underlying density function using asymmetric kernels. We assess the properties of these estimators with analytic results and simulation. We use actual images to evaluate the performance of our proposal.