Department of Computer Sciences, University of Wisconsin-Madison, 1210 West Dayton Street, Madison, WI 53706, USA.
IEEE Trans Vis Comput Graph. 2013 Sep;19(9):1526-38. doi: 10.1109/TVCG.2013.65.
We introduce Splatterplots, a novel presentation of scattered data that enables visualizations that scale beyond standard scatter plots. Traditional scatter plots suffer from overdraw (overlapping glyphs) as the number of points per unit area increases. Overdraw obscures outliers, hides data distributions, and makes the relationship among subgroups of the data difficult to discern. To address these issues, Splatterplots abstract away information such that the density of data shown in any unit of screen space is bounded, while allowing continuous zoom to reveal abstracted details. Abstraction automatically groups dense data points into contours and samples remaining points. We combine techniques for abstraction with perceptually based color blending to reveal the relationship between data subgroups. The resulting visualizations represent the dense regions of each subgroup of the data set as smooth closed shapes and show representative outliers explicitly. We present techniques that leverage the GPU for Splatterplot computation and rendering, enabling interaction with massive data sets. We show how Splatterplots can be an effective alternative to traditional methods of displaying scatter data communicating data trends, outliers, and data set relationships much like traditional scatter plots, but scaling to data sets of higher density and up to millions of points on the screen.
我们介绍了 Splatterplots，这是一种新颖的散点数据表示方式，能够实现超越标准散点图的可视化效果。传统的散点图在每个单位面积的点数增加时会出现过度绘制（重叠的图形）。过度绘制会使离群值变得模糊，隐藏数据分布，并使数据子组之间的关系难以辨别。为了解决这些问题，Splatterplots 抽象了信息，使得在任何屏幕空间单位中显示的数据密度都受到限制，同时允许连续缩放以揭示抽象的细节。抽象自动将密集的数据点分组为轮廓，并对其余的点进行采样。我们将抽象技术与基于感知的颜色混合相结合，以揭示数据子组之间的关系。生成的可视化效果将数据集的每个子组的密集区域表示为平滑的闭合形状，并显式显示代表离群值的点。我们展示了利用 GPU 进行 Splatterplot 计算和渲染的技术，从而能够与大规模数据集进行交互。我们展示了 Splatterplots 如何成为显示散点数据的传统方法的有效替代方法，能够像传统散点图一样传达数据趋势、离群值和数据集关系，但可以扩展到更高密度的数据集和屏幕上的数百万个点。
