SCANNING through large amounts of images or video footage can be time-consuming and laborious. Now a machine that taps into the processing power of your brain can help you do it much faster.
The device, called the cortically coupled computer vision system (C3Vision), was developed by Paul Sajda and colleagues at Columbia University in New York. It uses an electroencephalogram (EEG) to measure electrical activity in the brain via skull-cap electrodes, while the user scans through images. If the user sees an image that grabs their attention, such as a picture of a crime suspect, the EEG will pick up a spike of electrical activity in their brain 300 milliseconds later, something known as a P300 response. This occurs well before the user can click a button to say they saw the picture. A connected computer can then collect all the images associated with a P300 response.
Sajda says the system would be useful for intelligence analysts searching through images or video footage of a crime scene, or doctors scanning through medical imagery. 鈥淚t鈥檚 kind of like finding a needle in a haystack,鈥� he says. 鈥淚f you can quickly rummage through the haystack, you might actually spot something that gleams out and then you can more closely look at it.鈥�
Advertisement
Sadja says computer vision algorithms could be added to further speed up the process. For example, a computer could prune out images taken during the night, leaving a person to perform the trickier task of identifying a suspect in the remaining images.
Leif Finkel from the University of Pennsylvania in Philadelphia believes the system could prove a boon for information analysts in years to come. 鈥淭he balance between data volume and analysis time is only going to get worse as the amount and variety of information acquisition grows,鈥� he says.
Sajda鈥檚 team tested the device by showing volunteers 5000 images of outdoor scenes for 100 milliseconds each, one straight after another. A handful of these images included people, and the volunteers were asked to click a mouse button as soon as they spotted them. The speed and accuracy of their responses were compared with those of the C3Vision machine.
The machine produced faster responses, but it did sometimes produce false positives, for instance when a person鈥檚 attention was grabbed by something else. The most accurate method of finding images, the researchers found, was to combine the responses of both human and machine.
In a few months the machine will be demonstrated to DARPA, the US Department of Defense鈥檚 research arm, which has provided funding for the project.
