Multi-Scale DCT Denoising
Nicola Pierazzo, Jean-Michel Morel, Gabriele Facciolo
⚠ This is a preprint. It may change before it is accepted for publication.


DCT denoising is a classic low complexity method built in the JPEG compression norm. Once made translation invariant, this algorithm was still proven to be competitive at the beginning of this century. Since then, it has been outperformed by patch based methods, that are far more complex. This paper proposes a two step (oracular) multi-scale version of the algorithm that boosts its performance and reduces its artifacts. The multi-scale version decomposes first the image in a dyadic DCT pyramid, which keeps noise white at all scales. Single scale denoising is then applied to all scales, thus giving multiple denoised versions of the low frequency coefficients of the denoised image. A 'soft multi-scale fusion' of these multiple estimates avoids the ringing artifacts caused by hard frequency cut-offs. The final algorithm wins back a good PNSR and much improved visual image quality.