Multi-channel nuclear norm minus Frobenius norm minimization for color image denoising

Color image denoising is frequently encountered in various processing and computer vision tasks. One traditional strategy to convert the RGB a less correlated color space denoise each channel of new separately. However, such can not fully exploit information between channels inadequate obtain satisfactory results. To address this issue, paper proposes multi-channel optimization model for under nuclear norm minus Frobenius minimization framework. Specifically, based on block-matching, decomposed into overlapping patches. For patch, we stack its similar neighbors form corresponding patch matrix. The proposed performed matrix recover noise-free version. During recovery process, a) weight introduced utilize noise difference channels; b) singular values are shrunk adaptively without additionally assigning weights. With them, achieve promising results while keeping simplicity. solve model, an accurate effective algorithm built alternating direction method multipliers (ADMM) solution updating step be analytically expressed closed-from. Rigorous theoretical analysis proves that sequences generated by converge their respective stationary points. Experimental both synthetic real data sets demonstrate outperforms state-of-the-art models. MATLAB code available at https://www.github.com/wangzhi-swu/MCNNFNM.