VisualIDs: Scenery for Data Worlds

J.P. Lewis	Ruth Rosenholtz	Nickson Fong	Ulrich Neumann
CGIT Lab USC	Perceptual Science Group MIT	ESC Entertainment	IMSC USC

Search and memory for images is known to be generally faster and more robust than search and memory for words. The desktop metaphor was intended to allow us to engage these visual skills at the computer interface. Yet studies show that average computer users do not have a clear idea of where they save their data. This "lost in dataspace" problem has been addressed with a variety of focus+context approaches that add a small map or other indication of one's location to the GUI. But the same users who misplace files on a daily basis are not nearly as lost in the real world -- we do not need a map to find our way to the kitchen or locate our workplace each day. There are familiar places where people do get lost, however: big parking lots are one example.

The problem is obvious, though it has been largely overlooked in several decades of HCI research: scenery is what is missing. In current GUIs everything looks the same, and finding a particular file is as difficult as finding your car in a big parking garage. In this project we show that suitable scenery (distinctive appearance) can be automatically invented using computer graphics techniques. VisualIDs are a form of automatically generated persistent scenery applied to individual data objects such as files, for the purpose of providing faster and more effortless search over a working set of objects.

The VisualIDs approach makes use of the following (perhaps surprising) principle:

The appearance of an object does not need to be related to its content or meaning to the user.

Humans are quite capable of learning arbitrary appearance and associating it to objects. This is fortunate, because automatically identifying content can be difficult, and identifying its meaning to the user is arguably impossible in general (see FAQ).

Our experiments show significantly improved speeds in a simulated file finding task after only a few minutes of use, and with no training. Second-day accuracy in locating data objects is dramatically increased when the data icon includes a VisualID. Examples of possible application scenarios are:

Icons to identify objects in a monitoring situation such as air traffic control. Searching based on appearance is generally faster, more reliable, and less effortful than searching text (the text can be used to confirm the located object of course).
Icons for the set of files currently in use in a programmer's editor such as Visual C++. Instead of scanning names, scan the icons (again, visual search is faster than searching text).

FAQ

ACM Transactions on Graphics paper:
VisualIDs: Automatic Distinctive Icons for Desktop Interfaces

browser prototype screenshot

visualID varieties (1.5 meg image)

Supplementary material (pdf)

A Linux reference implementation and patches for the Nautilus file browser are under development by J. Rosen.