Perception-based Information Visualization

可视化像任何其他类型的图像一样为人所感知，因此选择合适的显示参数对于在可视化中忠实地传输底层数据至关重要。本项目旨在分析从计算机图形学、视觉和感知中已知的方法可以在多大程度上应用于判断和适当地创建信息可视化结果。在他们之前的工作中，申请人已经证明，例如，选择正确的纵横比可以从根本上改变对投影的高维数据的感知，并确定哪种可视化方法最适合于显示它。选择适当的纵横比对于渲染折线图和其他可视化效果也很重要。在这两项工作中，离散的可视化由密度图表示，这些密度图是通过使用高斯核的核密度估计(KDE)来创建的。在拟议的项目中，这一方法将扩展并研究哪些可视化类型可以使用哪种连续表示，以及如何通过自动手段对其进行分析以找到最佳显示参数。申请者将在几个方面开展工作：除了根据连续表示法选择适当的纵横比和其他图形属性外，他们还将检查功能图、常见的统计图形(如折线图或条形图)，并研究勾号和背景网格等装饰如何改变他们的感知。这些元素的良好分配可能有助于用户将注意力集中在数据的重要或有趣的方面。另一个方面是选择正确的颜色。颜色对比和平衡有助于优化选择和分类任务，在一项初步研究中，申请者证明了正确的颜色分配有助于在多聚类可视化中在视觉上区分聚类。在可视化的离散数据及其连续表示上，申请者希望应用感知规律来判断表现力。其中一位申请者(Deussen)在他之前的作品中展示了基于格式塔的定律可以用于分析图像和3D几何中的对象模式。这将应用于可视化的元素。此外，对于已经在计算机图形和视觉中使用的图像的分析，存在许多美学定律。申请人将使用它来优化可视化的视觉参数，同时他们将使用基础(离散)数据来优化在连续表示中看不到的方面，例如离群值、非常稀疏的填充区域。通过组合这两种表示的各方面，将估计出感知上最优的可视化参数。因此，申请人希望通过联合调查影响可视化的连续和离散因素来为感知驱动的信息可视化奠定基础。

Visualizations are perceived by humans like any other type of images, thus it is crucial to select appropriate display parameters for faithfully transporting the underlying data in a visualization. This project aims to analyse to what extent methods known from computer graphics, vision and perception can be applied to judge and appropriately create information visualization results. In their previous works the applicants have shown that, e.g., selecting the right aspect ratio can fundamentally alter the perception of projected high-dimensional data and determine which visualization method is best for showing it. Choosing an appropriate aspect ratio is also important for rendering line graphs and other visualizations. In both works discrete visualizations were represented by density maps that were created by using Kernel Density Estimation (KDE) with Gaussian kernels. In the proposed project this approach will be extended and investigated for which visualization types which kind of continuous representations can be used and how to analyse them by automatic means in order to find optimal display parameters. The applicants will work on several aspects: besides selecting appropriate aspect ratios and other graphical attributes based on continuous representations, they will examine functional plots, common statistical graphics (such as lines charts or bar charts) and study how decorations such as tick marks and background grids alter their perception. A good assignment of such elements might help users to focus on important or interesting aspects of the data. Another aspect is selecting the right colors. Color contrasts and balances help to optimize selection and classification tasks, in a preliminary study the applicants demonstrate that the right color assignment helps in visually distinguishing clusters in a multi-cluster visualization.On both, the discrete data of a visualization and its continuous representation, the applicants want to apply perceptual laws for judging expressiveness. One of the applicants (Deussen) has shown in his previous works that Gestalt-based laws can be used for analyzing object patterns in images and 3d geometry. This will now be applied to the elements of visualizations. Furthermore, a number of aesthetics laws exist for the analysis of images that have been used in computer graphics and vision. The applicants will use this to optimize visual parameters of visualizations and at the same time they will use the underlying (discrete) data to optimize aspects that cannot be seen in the continuous representation, such as outliers, very sparsely populated areas. By combining aspects of both representations perceptually optimal visualization parameters will be estimated. Thus the applicants want to lay the foundations of perceptually-driven information visualization by jointly investigating continuous and discrete factors that influence how visualizations are perceived.