= Menu

Improving Color for Data Visualization

PhD thesis from University of Wisconsin-Madison — July 2015
    Download the publication : dissertation.pdf [76.9Mo]  

    Visualization allows viewers to explore large collections of data. Effective visualizations must support viewers in understanding data both at high-levels to investigate “big picture” statistics, patterns, and trends, and at low-levels to examine individual values. Visualization design guidelines currently focus on how designs can support low-level tasks, such as determining if one value is larger than another, but far less is known about designing for high-level tasks. High-level tasks require viewers to aggregate information across multiple datapoints, such as estimating the average value of a set of points. Systems can explicitly compute these values, but must know the task and data that viewers are interested in in advance to do so. Instead, viewers frequently need to visually aggregate information across multiple datapoints in a visualization. However, designs that are effective for low-level tasks may not support visual aggregation, especially as datasets increase in size and complexity. To remain effective at scale, visualizations must consider how designs can support estimates both across multiple values (visual aggregation tasks) and between individual values (low-level tasks).

    This dissertation describes a set of experiments, metrics, and techniques that allow visualizations to more effectively support both high- and low-level tasks by using color. To support high-level tasks, I identify limitations that inhibit visual aggregation in existing visualization designs and introduce novel designs using color to overcome these limitations. I show how different decisions made in creating a visualization can support visual aggregation. I embody these results in visualization systems that increase the size of datasets analysts can explore for three different domains. I address challenges of using color for low-level tasks by generating metrics and guidelines for color encoding design tailored to visualization. I first show how visualization designs can improve perceptions of shadowed colors in surface visualizations. I then model how two factors of visualization viewing (viewing enviroment and mark size) influence color encoding perceptions in practice and show how these models can be used to guide effective encoding design. The main technical contributions of this dissertation include a method for task-driven aggregation for one-dimensional data, novel visualization systems for analyzing data in genomics, text analysis, and structural biology, and a data-driven method for modeling perceived color differences.

    Images and movies


    BibTex references

      author       = "Albers Szafir, Danielle",
      title        = "Improving Color for Data Visualization",
      school       = "University of Wisconsin-Madison",
      month        = "July",
      year         = "2015",
      url          = "http://graphics.cs.wisc.edu/Papers/2015/Alb15"

    Other publications in the database