An automatic additive and multiplicative noise removal scheme with sharpness preservation

To remove noise from biomedical images polluted by excessive and inhomogeneous additive or multiplicative noise, most of the denoising algorithms cannot keep a desirable balance between denoising and preservation of fine features; only work for one specific noise; and involve heuristic parameter tuning. We present a fully automatic approach to preserve sharp edges and fine details while removing noise. Explained in nonlocal means scheme, we propose a segmentation boosted NL-means filter (SNL) based on the concept of mutual position function to ensure averaging is only taken over pixels in the same phase. To address unreliable segmentation due to excessive noise, we apply SNL filtering in an iterative way. Comparison with ROF, BM3D, K-SVD and the original NL-means on simulated data, MRI and SEM images indicates potentials of our method.