A method for predicting denoising efficiency for color images

A method to predict denoising efficiency of filters based on discrete cosine transform (DCT) for multichannel images is proposed. Multichannel (3D) standard DCT-based filter and C-BM3D are considered with application to suppression of additive white Gaussian Noise (AWGN) in RGB-color images. PSNR and PSNR-HMA metrics are exploited to assess quality of images before and after filtering. The proposed method uses image statistics in DCT domain and fitting simple monotonic functions into scatter-plots for prediction of filtering efficiency. It is shown that the prediction method has high performance and requires computational burden much lower than DCT-based filtering itself.