Alpha, contrast and the perception of visual metadata
Alpha, contrast and the perception of visual metadata
复制标题
Alpha、对比度和视觉元数据的感知
DOI：
--
复制
发表时间：
2008
期刊：
影响因子：
--
通讯作者：
L. Bartram
中科院分区：
文献类型：
--
作者：
M. Stone;L. Bartram
网格、标签和等高线等视觉元素充当支持所呈现的主要信息的“参考结构”或“视觉元数据”。这样的结构需要有用地可见，但又不能太过突出，以至于扰乱了演示文稿。我们的目标是确定这类结构的物理、感知和认知特征，理想情况下是以一种能够自动计算的方式。我们给出了一组实验的结果，以确定散点图数据上的薄矩形网格的有效显示范围，以透明度(Alpha)的形式描述。这些结果表明，有效范围可以用阿尔法来定义。为了创建一组独立于显示的指标，我们从亮度对比度的角度分析了这些结果，结果喜忧参半。我们的结论是，透明度的外观是微妙可视化的一个重要方面。简介网格等可视化元素充当支持所呈现的主要信息的引用结构或可视化元数据。这样的结构需要有用地可见，但又不能太过突出，以至于扰乱了演示文稿。其他静态示例包括标签和等高线。智能光标和对象句柄等交互技术也创建了引用结构。视觉设计师熟练地处理颜色、线宽和透明度等属性，以在参考结构和关键数据之间创造平衡。我们研究的广泛目标是创建工程指标和模型，使动态的、算法生成的显示具有类似的有效性。我们处理这个问题的方法不是描述“理想的”或“最佳的”，而是定义边界条件，超出边界条件的结果显然是糟糕的。我们的理由是，最佳解决方案总是会受到背景和品味的影响。然而，边界条件更有可能具有简单的规则，工程师和研究人员可以很容易地将其纳入其中。通过消除或至少减少最令人反感的情况，我们可以更容易地提高计算机生成演示文稿的整体质量。本文将总结第一组实验的结果，以表征矩形网格的边界条件。在这些实验中，受试者操纵了覆盖在不同复杂程度的散点图上的细线网格的透明度(Alpha)，这些散点图呈现在不同亮度的背景上(均为灰度图像)。我们的目标是找到一系列Alpha值，以创建可以接受的细微网格。我们的结果表明，对于我们的实验条件，可以建立统计上可接受的阿尔法值范围，该范围对所有受试者使用相同的校准显示。要创建独立于显示的模型，我们需要将结果与感知指标(如亮度对比度)绑定，例如，亮度对比度用于提供文本易读性指标。我们的分析表明，仅有反差不足以解释我们的结果，这表明透明度可能是更关键的衡量标准。本文将首先从设计的角度讨论参考结构，然后总结我们的实验及其结果。然后我们从对比的角度对我们的结果进行分析。最后，我们总结了我们未来工作的方向，重点是图像复杂性和对透明度度量的进一步探索。微妙设计设计师通过不同的视觉对比度来创建微妙的参考结构，通常是操纵颜色、线宽和透明度[1]。图1显示了覆盖在地图上的网格。图1(B)中定义网格的线条看起来比图1(A)中的线条更亮(实际上更透明)，也更细，从而产生更微妙的外观。设计师的总体目标是实现视觉层次的良好平衡的组成，其中任何组成“图形”的东西都相对于“背景”被很好地定义。网格和其他可视元数据位于这些层中间的某个位置，有时网格需要更多的图形(便于搜索或参考)，有时需要更多的地面(降级到背景中，而不是侵入性)。图1(A)。设计糟糕的网格模糊了底层信息。图1(B)。这个网格更细微，允许观众聚焦
Visual elements such as grids, labels, and contour lines act as “reference structures” or “visual metadata” that support the primary information being presented. Such structures need to be usefully visible, but not so obtrusive that they clutter the presentation. Our goal is to determine the physical, perceptual and cognitive characteristics of such structures, ideally in a way that enables their automatic computation. We present the result of a set of experiments to determine effective display ranges, described in terms of transparency (alpha), for thin rectangular grids over scatterplot data. These show that an effective range can be defined in terms of alpha. In an effort to create a display-independent set of metrics, we analyze these results in terms of luminance contrast, with mixed results. We conclude that the appearance of transparency is an important aspect of subtle visualization. Introduction Visual elements such as grids act as reference structures or visual metadata that support the primary information being presented. Such structures need to be usefully visible, but not so obtrusive that they clutter the presentation. Other static examples include labels and contour lines. Interactive techniques like smart cursors and object handles also create reference structures. Visual designers expertly manipulate properties such as color, line weight and transparency to create a balance between reference structures and the critical data. The broad goal of our research is to create engineering metrics and models that enable dynamic, algorithmically generated displays to be similarly effective. Our approach to this problem is not to characterize “ideal” or “best,” but instead to define boundary conditions, outside of which the result is clearly bad. We reason that the best solution will always be influenced by both context and taste. Boundary conditions, however, are more likely to have simple rules that can easily be incorporated by engineers and researchers. By eliminating, or at least reducing, the most objectionable cases, we can more easily raise the overall quality of computer-generated presentations. This paper will summarize the results from a first set of experiments to characterize the boundary conditions for rectangular grids. In these experiments, the subjects manipulated the transparency (alpha) of thin-line grids overlaid on scatterplots of different complexities rendered on backgrounds of different lightnesses (all grayscale imagery). The goal was to find a range of alpha values that create acceptably subtle grids. Our results show that a statistically acceptable range of alpha values can be established for our experimental conditions, which used the same calibrated display for all subjects. To create a display-independent model, we need to tie our results to perceptual metrics such as luminance contrast, which is used, for example, to provide metrics for text legibility. Our analysis shows that contrast alone is insufficient to explain our results, suggesting that the degree of transparency may be the more critical metric. This paper will first discuss reference structures from a design perspective, then summarize our experiments and their results. We then provide an analysis of our results in terms of contrast. We conclude with our directions for future work, focusing on image complexity and further explorations of transparency metrics. Subtle Design Designers create subtle reference structures by vary visual contrast, typically manipulating color, line weight and transparency [1]. Figure 1 shows a grid overlaid on a map. The lines that define the grid in Figure 1(b) appear lighter (actually, more transparent) and are thinner than those in Figure 1(a), resulting in a more subtle appearance. The overall goal of the designer is to achieve a well-balanced composition of visual layers, in which whatever constitutes the “figure” is well defined with respect to “ground”. Grids and other visual metadata live somewhere in the middle of these layers, where sometimes the grid needs to be more figure (visually accessible for search or reference) and sometimes more ground (relegated to the background and not intrusive.) Figure 1(a). A badly designed grid that obscures the underlying information. Figure 1 (b). This grid is more subtle, allowing the viewer to focus