Realtime Visualization of Large Images on a Thinwire

Problem Statement

We address the following issue: many visualization applications involve very large or complex images. Many visualization servers on the internet pre-select a set of subimages of this complex image for the visualizer. These subimages are of limited resolution and size (say 640x480) because of thinwire limitations. The pre-selected images are unlikely to be optimal for any particular user. Notice that the bandwidth usage in such servers highly asymetrical: practically all the bandwidth is from server to the user.

An improved approach is taken by map servers: users can now zoom and pan over the entire image. But current map servers suffer from three problems:

  1. visual discontinuities in zooming and panning (usually a brand new image is served up for each zoom/pan request)
  2. smallness of viewing window (e.g. 3"x4.5")
  3. distinctly non-realtime responses
How can we remove these constraints within current bandwidth limitations? We show how to effectively solve them using foveation methods.

Active Image Servers

We have constructed an active image server in which one can visualize images of unlimited size and resolution, without significant impact on the realtime response time. Here is a screen shot:

Notice that the main image is "multifoveated". The user can *continuously* zoom and pan over the image. The viewing area is freely resizable. The received images are multi-foveated, but the user can demand more detail at any spot of interest just by placing the mouse over the spot. The user can dynamically vary the foveation radius as well the rate of resolution decay. It might appear counter-intuitive that such capabilities is desirable in a thinwire setting. But the key observation is this: we are, in effect, drastically reducing the server-to-client bandwidth in exchange for a modest increase in the client-to-server bandwidth.

To access this server, you will need a corresponding client program. Please click here for information on obtaining this client. The above demo treats raster images. This technology has been patented and available for commercial license under the name Foveal Point. See our related GIS on the Web project where we address similar issues for vector data sets.

Related Links

Project Members: Ee-Chien Chang, Chee Yap, Ting-Jen Yen.