Randomized coherent sampling for reducing perceptual rendering error

Realistic image synthesis using path tracing needs many samples to achieve noise-free images. The noise is due to the use of Monte Carlo integration in path tracing. Since Monte Carlo integration evaluates each pixel using random sampling, we obtain noisy pixels at low sample counts. Due to the random nature of Monte Carlo integration, pixel values with finite numbers of samples can be significantly different, even if their correct solutions to the rendering equation are the same.