A New Dictionary Based on Local Mean for SAR

High-quality dictionaries contribute to reducing reconstruction time while maintaining reconstruction quality in image compression. Traditional dictionaries constructed from synthetic aperture radar (SAR) images using the K-singular value decomposition (K-SVD) algorithm often entail substantial computational cost during image reconstruction. Since the main features of SAR images can effectively characterize the overall image content, and the local mean of image blocks conveys important structural information, we propose a novel dictionary learning method that incorporates local mean information to preprocess training images. This approach preserves essential features of the training data, and the preprocessed images are then used to learn a dictionary via K-SVD. Image reconstruction is performed using the smoothed L0 norm at different measurement rates. Simulation results demonstrate that the proposed algorithm achieves improved performance in both computational efficiency and peak signal-to-noise ratio (PSNR).