A hybrid Hertzsprung-Russell diagram comparing the color index (x) and absolute magnitude (y) of the 30,000 stars closest to Earth. From close to the display, it is possible to examine the dataset at the level of individual stars.
To support far viewing, we add an aligned dual-scale grid that allows viewers at both near and far distances to read axis values and compare points. In addition to finer, more closely-spaced gridlines, the near image contains axis labels at the edge of each screen in the display, allowing close viewers to read values without having to look all the way to the edge of the wall.
We also tailor the far view to help viewers understand the global structure of the dataset and identify outliers. In the far image, we render stars using larger points that can be visually aggregated when viewed from several meters back. The far image's larger points are clearly visible from a 4m distance, but recede as the viewer approaches the wall and do not distract from the reading of the near-image star points, grid lines, and the star labels.
When plotting this many data points, important values like outliers can be easily missed. To address this, we also use the far image to highlight outliers by scaling the size of each point in the far image based on the distance to its nearest on-screen neighbor.
One important component of any text visualization is the accessibility of the underlying texts. When displaying raw text pages and viewing them from afar the typical 10-12 point font size makes it quickly impossible to analyze document structures or get an idea of the content of individual pages of text. Hybrid-image visualizations for large wall displays make it possible to render a large number of text pages and overlay them with visual summaries.
Here, we display multiple PDF documents side-by-side so that analysts can compare their structure and content. The far image highlights section titles and exposes the structure of the document. From a distance, section titles are clearly visible over the individually rendered text while from close by the title text is so blurred that it does not interfere with reading the actual words in the document. Conversely the text of the near image recedes when viewed from a distance.
This view allows viewers to identify similar structures and topics in the documents from a distance. The visual summaries of the far image can also serve as entry points of physical navigation and as aids for finding documents of interest.
This dual-scale network visualization shows the full co-authorship network of our research center. When close to the display, the visualization shows the full network, including about 3400 authors, 47 research groups, and 21300 links. Each author in the network is connected to another if they co-authored at least one publication together. Viewed from afar such a network graph becomes quickly unreadable as it consists mostly of small nodes that represent authors and narrow-width links that display connections between them.
To provide an overview of this dense graph, we display the simplified meta-structure of the network containing only the names of research groups and the connections between them. This layer is optimized to be seen from afar and is designed so that it will not visually interfere with the dense graph. The far visualization is visually aligned with the complete co-authorship network, allowing viewers to transition smoothly between the two views.
This treemap shows scientific classifications of living organisms in the animal kingdom (classifA data from the information visualization benchmark repository at http://www.cs.umd.edu/hcil/InfovisRepository/contest-2003/). The tree is very large, including 15 levels and 190,265 nodes. To create the image we rendered the borders for all nodes on the first four levels and labels for the first two levels into the far image at appropriate sizes so they can be see from far away from the wall. The near image contains the borders for all lower-level nodes and labels for all leaves where possible. As the color coding was important to give a sense of the depth of a node, we rendered filled nodes into a third background layer to avoid distortion due to blending and frequency filtering. The borders and labels of higher level nodes overlap deeper nodes but, due to the blur and blending, this overlay does not interfere with their readability from closer to the display.
Near and far images can also contain very different representations of the data. This visualization demonstrates mixing multiple visualizations by overlaying two different renderings of historical temperature data from 32 European cities (from http://www.ecad.eu). From a distance, viewers see small multiple bar charts showing the difference between the monthly average temperature for the most recent year and the historical average for that month between 1990 and 2012. However, from closer to the display, the bar charts become less salient and are replaced by small time series showing the individual daily temperature readings in that city for every day of every year.
Viewers can examine the bar charts from a distance to identify larger temperature trends and identify cities where the last year's temperatures are abnormal, then move closer to inspect and compare the raw data. Using a representation with mostly low spatial frequencies (bar charts) for the far image and one with mostly high frequencies (small line charts) for the near image minimizes the inter- ference between the two.