Abstract

Denoising of path-traced images, generated by multi-dimensional Monte Carlo integration, has a great potential in the area of image synthesis. This project focuses on design and development of novel algorithms for real-time denoising of synthetic images on mobile platforms. The proposed algorithms operate in spatial and temporal domains to provide maximum efficiency while keeping the quality of the image high. Our algorithms firstly filter the rendered samples in the temporal domain using reprojection and sample validation. Then, a fast spatial filter is applied after variance estimation to obtain final high-quality denoised image. The rendered samples contain also auxiliary information, including positions, normals, and secondary hit information to further drive the adaptive denoising filter. This research project is conducted in collaboration with HUAWEI Technologies CO, Ltd. Company.

Project partners

  • HUAWEI Technologies CO, Ltd.  (Shenzhen China)

Funding provided by

  • HUAWEI Technologies CO, Ltd.