CRII: CHS: Data-Driven Automation of Color Encodings for Data Visualization

University Of Colorado At Boulder, Boulder CO

Investigators

Abstract

Graphs, charts, and other visualizations of data rely on color both to convey key aspects of the underlying data and to attract and engage viewers. Getting both the accuracy and aesthetics of color choices right, however, is hard, and most existing tools for helping designers focus on just one of the two. Developing accurate color mappings is even harder because how colors are perceived changes depending on the size and shape of visual marks, lighting and contrast, and a number of other factors. In this project, the research team will use designs created by existing tools to construct an initial statistical model of color mappings that captures expert designers' current decision-making. They will then improve those models by creating visualizations based on the models, altering size, shape, contrast, and lighting, and testing how well people can use those designs to learn the underlying values of the data. Finally, the team will create a design tool that allows both expert and non-expert designers to create visualizations, choosing anchor colors and aspects of the visualization, and generating color maps that are most accurate and aesthetic based on the models and the designer's choices. The work will lead to more accurate models of perception and mechanisms for choosing color maps that capture both design expertise and perceptual accuracy; this, in turn, will lead to practical improvements in the effectiveness of data visualizations that are increasingly part of people's experience. The team also plans to increase the accessibility of data visualizations by helping designers choose color mappings that are more usable by people with color-blindness, while making the tools themselves more usable by color-blind people. The tools and work will also be integrated into several courses on human-computer interaction and data science at the lead investigator's institution, benefiting students from a variety of research groups and departments. Color ramps will be represented as a set of control points (two end points in sequential encodings and two end points plus a midpoint in diverging ramps) that determine the overall structure of the ramp, and a smooth interpolation path that connecting the control points in colorspace. To capture current expert practice, the team will first extract initial color ramps from colormaps available in existing design-based visualization tools, using the CIELAB colorspace to model the statistical characteristics of the control points and interpolation paths of these encodings, generating aesthetic constraints grounded in the current design consensus. The team will then use crowdsourcing platforms, which have been shown to be effective for a number of perceptual and visualization experiments, to systematically study how specific aspects of visualization design including mark shape, mark size, and visualization type, affect people's ability to detect color differences in colorspace; further, conducting the experiment online means this model will be specifically tailored to the online/web/screen viewing context. This empirical model can enforce perceptual constraints imposed by visualization design choices on the color ramps generated by the aesthetic models by constraining and repositioning control points. Finally, these models will be integrated into a publicly available color authoring system that will be validated through use in courses at the lead researcher's institution and at design workshops with the local community. In addition to developing the specific models and tools around color encodings, the work sets up a broader research agenda of combining automation and interaction, in which semi-automated guidance democratizes effective visualization practice and allows people to leverage prior designs and create new representations without requiring extensive visualization training.