Visual preference of blue noise in path tracing

Abstract

Blue noise is generally considered as preferable to other types of noise in graphics. However, there is a lack of empirical evidence for this claim. With this research, we aim to test this claim and provide additional understanding of the subject. We discuss literature on blue noise, sampling theory, and human vision to explain possible reasons for the preference of blue noise. We have performed an experiment to test whether rendered images with error distributed as blue noise were found to be visually preferable to images with error distributed as white noise. We found that there is a clear preference for blue noise over white noise when the amount of samples per pixel is equal. We also found that a preferable noise type does not outweigh the visual improvement of rendering an image with two samples instead of one.