The human visual system implements an elegant comprise between the competing goals of maximizing field of view, maximizing spatial resolution, and minimizing neural resources: It encodes a large field of view using a retina with variable spatial resolution, and then when necessary, uses high-speed eye movements to direct the highest-resolution region of the retina (the fovea) at specific points in the visual scene. For a number of years our laboratories both singly, and in collaboration, have been exploring the computational, biological and engineering implications of space variant visual systems, communication systems, and display systems.

One form of space variant imaging is foveated imaging, where the spatial resolution of the encoded or displayed image varies in a way that corresponds to the spatial resolution of the human visual system (as described above). For example, the image below shows a frame from a video sequence, where the spatial resolution is falling off smoothly from the point of gaze (the plus sign) on that particular frame. Bovik’s lab has explored the use of foveated image processing in robotic active vision systems. All three labs have explored various aspects of using foveated imaging for data compression and low-bandwidth video communications. Recently, Geisler’s lab has focused on the development of real-time algorithms and software for creating arbitrary space-variant video displays; these systems are being used in conjunction with eye tracking to precisely control the information across the visual field, while observers are performing complex tasks that involve multiple eye fixations.

For moreinformation, demonstrations and down-loadable software see:

CPS Space Variant Imaging
Laboratory for Image and Video Engineering
Laboratory for Artificial Neural Systems