SAR image despeckling via spatially constrained multiscale products

This paper presents a wavelet-based multiscale products scheme for synthetic aperture radar (SAR) image despeckling. A compactly supported quadratic spline function that approximates the first derivative of Gaussian is employed in the scheme to decompose log-transformed SAR images. The multiplied results of the decomposed coefficients of adjacent scales consist of multiscale products. The multiscale products can sharp the important structures while weakening noise. A spatially selective neighborhood technique by iteratively selecting neighborhood system in the multiscale products is introduced in searching the important structure information. The influence of the spatial information is imposed on the multiscale products, instead of on the wavelet coefficients, which improves the capability of identifying important features. Experiments show that the proposed scheme is better in SAR image despeckling and preserving edges and detail information than other wavelet-based multiscale products methods.