Field of experts regularized nonlocal low rank matrix approximation for image denoising

The restoration of image degraded by noise is an essential preprocessing step for various imaging technologies. Nonlocal low rank matrix approximation has been successfully applied to image denoising due to the capability of recovering the underlying low rank structures. Unfortunately, existing rank minimization models ignore the correlation among image patches and their performance is degraded when encountering the heavy noise. To address this, we propose a field of experts regularized nonlocal low rank matrix approximation (RFoE) denoising model, which integrates a global field of experts (FoE) regularization, a fidelity term, and a nonlocal low rank constraint into a unified framework. The weighted nuclear norm is adopted as the low rank constraint while the FoE prior is utilized to capture the global structure information. An alternating direction minimization algorithm based on half quadratic splitting can effectively solve this model. Extensive experimental results demonstrate that the proposed RFoE model has a superior performance.