Visual processing	Mathematical modeling

James Brakefield, a senior research engineer at KRUG, explained that vision researchers use a model-based approach to scientific experimentation when implementing a model, simulating experiments and determining whether experimental designs will work as intended. Later, experiments are conducted using test subjects. The actual results are then compared with simulated results.

KRUG uses the Wilson model, a mathematical model of early vision that attempts to define the visual cortex, the part of the brain that processes visual signals. Although the model, based on a 1989 paper by Hugh Wilson, is accepted as one of the most accurate available, it is compute-intensive, involving huge data sets that require large amounts of computer memory. KRUG researchers treat parts of the model as image-processing sequences, using them for experiment simulation. PV-WAVE, Visual Numerics' visual data analysis (VDA) software, completes these equations in minutes.

"PV-WAVE makes implementing the model-based approach to science possible," said Brakefield. "It improves the quality of science by making the scientific process more interactive, bringing the scientist back into the engineering loop, reducing the number of people involved," he continued. "It brings big scientific projects to a more personal level, allowing more interaction with the process."

KRUG scientists have developed several model-based experiments to help them research vision. PV-WAVE is used during various segments of the project. Some of the experiments are discussed here.

Data reduction projects

Digitized Eye Movement

Test subjects track a cursor across a CRT screen. The cursor and eye-position trajectories are saved and later examined using PV-WAVE.

Digitized Foveal Images

The back of the subject's eye is videotaped. PV-WAVE analyzes the digitized video by tracing the frame-by-frame changes, which are fitted to an exponential curve.

Digitized Behavioral Data

The performance of a subject is tracked over time. The data are read into a menu-driven PV-WAVE application for analysis and reduction. Six graphs and associated measures are produced. The application allows control of end-points and runs on a standard graphics terminal.

Visualization projects

Loss of Image Recognition

Prototypical images are digitized with a camera and then filtered to simulate various types and amounts of visual deficit. This includes blurring, reduction of contrast and addition of after-images.

Visual Deficit Dynamics

• The noise level of a 16-frame file loop is adjusted along with the contrast and frame rate. The visibility of a target is then studied.
• The decay of an after-image is done in real time. The target appearance time can then be compared with experimental data.
• PV-WAVE is used to precompute a series of increasingly blurred images. These are then combined in real time to create a composite image representative of human vision. The center of vision is adjusted using a mouse. A variable transport delay can be included such that the center of vision as displayed on the frame buffer lags the mouse. This gives some idea of the effect of transport delay on visual perception.

Before purchasing PV-WAVE, Brakefield used home-grown data-analysis programs. Because of the software's speed and interactive nature, Brakefield estimates that he is five times more productive since he began using the product. PV-WAVE has simplified prototype implementation at KRUG, which lets researchers determine within minutes whether experimental ideas are viable.

"With PV-WAVE, we are able to add other devices such as frame buffers, and we can incorporate code from other sources. It provides interfaces that are efficient and easy to use, but it still can be customized to suit our needs," Brakefield says. "I've been extremely happy with PV-WAVE. I can finally program a 32-bit computer the way I've always wanted to," he concluded.