A new adaptive windowing method for SAR image despeckling filters

Abstract

Filter window selection is one of the key issues in SAR image despeckling. This paper proposes an adaptive windowing method for robust estimation of the local statistics for despeckling filters, which is based on the combination of confidence interval and morphological reconstruction. A preliminary homogeneous window of each pixel is firstly shaped from an initial window according to the confidence interval inferred by a given confidence probability. The confidence probability is chosen adaptively according to the homogeneity facts of the initial window. Subsequently, this preliminary window is refined by morphological reconstruction under the region adjacency constraints. As a result, a homogeneous window for filtering with arbitrary shape is obtained, which is continuous in both radiometric and spatial domain. The experimental results show that the proposed adaptive windowing method performs better in the term of the window accuracy and gets better balance between speckle reduction and structure preservation with two other commonly used windowing method.