SAR Image Formation and Feature Separation via Low Rank Sparse Decomposition

In this paper, a novel synthetic aperture radar (SAR) image reconstruction method is introduced that enhances and inherently separates the features of interest in the scene. Conventional SAR image formation methods have no mechanism to deal with regions with non-smooth/non-sparse features and may produce images which are not well-suited for interpretation tasks such as segmentation and automatic target recognition. These regions may be represented as low-rank structures, and can be tackled by formulating the SAR image reconstruction as a low rank plus dictionary-based sparse decomposition (LRDSD) problem. Proposed method can be applied broadly to the scenes that exhibit low rank property together with the sparsity in associated with any dictionary. Therefore, through LRDSD a SAR image is jointly reconstructed with enhanced features, and decomposed into its low-rank background and the feature of interest. An algorithm is proposed to solve the relevant convex optimization problem based on the alternating direction method of multipliers (ADMM). The method is validated with numerical results and we show the effectiveness of the proposed method on both synthetic and real SAR images.