Denoising in Monte Carlo rendering based on Clustered-Residual block network

Monte Carlo rendering image denoising is different from natural image denoising. The auxiliary information can be used to reconstruct the image. However, ordinary convolutional networks cannot handle MC noise well, and it is difficult to establish a suitable model between noise, auxiliary information and the original image. In this paper we addresses the above issues, propose a Monte Carlo image denoising network based on deep neural network, construct a residual aggregation module, add dense connections, and extract image detail features. Combining the three most commonly used auxiliary information (depth, normal, and reflectivity) helps the network accurately restore the image. The experimental results show that the algorithm proposed in this paper can adapt to various typical scenes and effectively reduce the noise of rendered images.