Plug-and-Play Priors Enabled SAR Image Inpainting in the Presence of Speckle Noise

Abstract

Synthetic aperture radar (SAR), being a coherent imaging system, usually produces images that are affected by granular deformities known as speckle. Image restoration from such noisy observation is an ill-posed problem. Model-based optimization is the framework that effectively tackle such inverse problems by building the degradation model and utilizing the prior information. The modular structure of alternating direction method of multipliers (ADMM) converges to solution by iteratively minimizing the cost function, which is the sum of the above two models. Recently, Plug-and-Play (PnP) ADMM is developed, which provides the flexibility to use image denoisers in place of regularizers. Image estimation from partial/incomplete observation is quite challenging and open topic of research in the literature. In this paper, SAR image reconstruction under multiplicative noise is discussed for image inpainting using PnP ADMM. Simulation results show that denoisers can be used to restore the images affected by large fractions of missing pixels.