Abstract:Classical vectorbased or matrixbased denoising methods for remote sensing images (RSIs) may cause distortion both in spatial domain and spectral domain. To improve denoising performance and reduce the distortion, a denoising method based on multi-linear weighted nuclear norm is proposed. Firstly, by considering spectral continuity and interdependency of three unfolding matrices along three modes of RSI, a multi-linear rank is constructed to model the spatial and spectral nonlocal similarity. Then, to make the proposed method more tractable, a variable splittingbased technique is used to solve the optimization problem. Experiment results reveal that the proposed method can provide substantial improvements over the current stateoftheart methods in terms of both objective metric and subjective visual quality.