Real-Time Filtering of Monte Carlo Noise on GPU

Master Thesis

About This Topic

Computational capabilities of modern graphics hardware enable physically based rendering algorithms to run in real-time using low sampling rate. The problem of insufficient sampling rate is the noise produced by random parameters in Monte Carlo integration. Several advanced algorithms for filtering and multidimensional reconstruction have been presented to address this problem. The main goal of this thesis will be to improve existing filtering algorithms and create a GPU implementation of selected algorithm. Additionally, novel algorithms can be derived. Developed filtering approach will be joined with a real-time GPU path tracer to evaluate the final results.