Deep SNR-estimation for Wiener filter-based non-blind deconvolution

We propose a data-driven approach for estimating the spatial frequency-dependent signal-to-noise ratio based on an observed image to achieve competitive deconvolution results with the classical Wiener filter. Despite its simplicity, the proposed approach is on par with, or even outperforms, the state of the art that relies on sophisticated models. The reconstructed images are of high quality and only exhibit weak artifacts mostly visible in image regions with low texture. Additionally, the simple approach enables fast reconstruction speeds.