Denoising of astrophysical image mixtures from spatially varying noise

In this study, some denoising methods are compared for denoising astrophysical image mixtures from space varying noise and a graph lowpass filter based vertex-frequency graph Wiener filter is proposed for this case of space varying noise power. Performances of the methods in terms of peak signal-to-noise ratio (PSNR) are investigated on image mixtures consisting of cosmic microwave background radiation (CMB), sychrotron and galactic dust, simulated to be obtained through 70 GHz and 100 GHz microwave channels, for different mixture coefficients. Our proposed graph Wiener filter algorithm is compared with five other prominent denoising methods; nonlocal means improved by the Sinkhorn algorithm (NLM+S), BM3D, NLGBT, OGLR and a Wiener filter algorithm in the pixel domain. Our algorithm is slightly surpassed by the NLM+S algorithm in all of tried six mixing scenarios but generally gives competitive results.