A VARIATIONAL MODEL TO REMOVE MULTIPLICATIVE NOISE BASED ON SAR IMAGE FEATURE PRESERVATION

Multiplicative noise frequently occurs in synthetic aperture radar (SAR), significantly deteriorating the image's feature information and posing challenges for subsequent analysis. Considering that SAR images typically exhibit a high-intensity range, we propose a non-convex variational multiplicative noise removal model that can be applied to images of any intensity range. In this work, SAR images with unknown noise levels are normalized to preserve their full content. To protect image features effectively, we introduce a new degradation model with nonlinear transformation. With the maximum a posteriori (MAP) estimator, we establish a new fidelity term to preserve and enhance the image feature. To address the staircase effect, an adaptive regularization term is introduced. Although the proposed model is non-convex, we prove the existence and uniqueness of the solution in some case. The model is solved using the finite difference method (FDM) with rescaling techniques and the primal-dual method (PDM). The proposed model can preserve and enhance the features of the image compared to the over-smoothed results obtained by the deep learning model and the results obtained by traditional PDE models.