Is it darker? Improving density representation in 2D scatter plots through a user study

Density differences are one of the main features users perceive in 2D scatter plots. However, because of pixels' collisions, some areas become saturated and such differences are lost. To solve this problem, several proposals rely on sampling the dataset before visualizing it. Some of these introduce precise measures to understand the image degradation and use numerical differences in pixels to estimate density differences. It is our opinion that this issue deserves a deeper analysis, taking into account perceptual issues. In this paper we describe a study we conducted to understand the relationship between numerical pixel density and the perceived density. The results obtained were used to refine a sampling technique we developed to preserve relative densities in the context of 2D scatter plots.