Data-Driven Colormap Optimization for 2D Scalar Field Visualization
Qiong Zeng1     YinqiaoWang1     Jian Zhang2     Wenting Zhang1     Changhe Tu1     Ivan Viola3     Yunhai Wang1
1Shandong University     2Computer Network Information Center, Chinese Academy of Sciences     3King Abdullah University of Science and Technology
Accepted by IEEE Vis 2019 Short Papers
Illustration of ocean salinity encoded with different colormap measures. (a) shows the input salinity rendered with an input Blue-White colormap. The salinity values range from 1 to 40 psu, with a few extreme low values in the Baltic Sea (pointed with a red arrow). (b) shows the visualization made by a domain expert, which ignores the extreme values and limits the visualization to range between 31 and 41 psu. (c) uses histogram equalization to adjust control points in the colormap according to cumulative data distribution frequencies. (d) is visualized with our automatically optimized colormap. Our optimization algorithm aims to reveal hidden data variations, with consideration of colormap originality and color visibility.
Colormapping is an effective and popular visual representation to analyze data patterns for 2D scalar fields. Scientists usually adopt a default colormap and adjust it to fit data in a trial-and-error process. Even though a few colormap design rules and measures are proposed, there is no automatic algorithm to directly optimize a default colormap for better revealing spatial patterns hidden in un- evenly distributed data, especially the boundary characteristics. To fill this gap, we conduct a pilot study with six domain experts and summarize three requirements for automated colormap adjustment. We formulate the colormap adjustment as a nonlinear constrained optimization problem, and develop an efficient GPU-based implementation accompanying with a few interactions. We demonstrate the usefulness of our method with two case studies.
This research was supported by the grants of NSFC (61602273, 61772315, 61861136012), Science Challenge Project (TZ2016002), and by the funding from King Abdullah University of Science and Technology (KAUST) under award number BAS/1/1680-01-01. This research used resources of the Core Labs of KAUST. The authors would also like to thank Kresimir Matkovic at VRVis Center for Virtual Reality and Visualisation GmbH (Vienna, Austria), Renata Raidou at TU Wien (Austria), Michael Böttinger at Deutsches Klimarechenzentrum (Germany), Thomas Theussl at KAUST (Saudi Arabia), Mingkui Li at Ocean University of China and Qianqian Guo at Shandong University (China) for providing precious visualization resources and evaluating the quality of our cases.