Anwar Saeed ; Barnes Nick

Abstract

Deep convolutional neural networks perform better on images containingspatially invariant noise (synthetic noise); however, their performance islimited on real-noisy photographs and requires multiple stage network modeling.To advance the practicability of denoising algorithms, this paper proposes anovel single-stage blind real image denoising network (RIDNet) by employing amodular architecture. We use a residual on the residual structure to ease theflow of low-frequency information and apply feature attention to exploit thechannel dependencies. Furthermore, the evaluation in terms of quantitativemetrics and visual quality on three synthetic and four real noisy datasetsagainst 19 state-of-the-art algorithms demonstrate the superiority of ourRIDNet.