Choice of the Parameter for BM3D Denoising Algorithm Using No- Reference Metric

An automatic multiscale algorithm for Block-matching and 3D filtering (BM3D) method de noising parameter selection has been proposed. To optimize the filtering parameter the presence of retained structures in the ridge areas is analysed for the difference of the initial noisy and filtered images. Appearance of regular components on method noise is controlled using mutual information. An estimation of image characteristic details is based on Hessian matrix eigenvalues analysis. Images with added controlled Gaussian noise from general image TID2013 and BSDS500 databases were used for testing. It was found that the proposed no-reference metric outperforms existing no-reference metrics in selecting optimal denoising parameter. Algorithm calculation time does not depend on the image noise level and looks promising to be used in image adaptive BM3D-based methods.