Training Humans to Spot Abnormalities in 50 Milliseconds
After looking for just one-twentieth of a second, experts in camouflage breaking can accurately detect not only that something is hidden in a scene, but precisely identify the camouflaged target, with great potential in medical diagnostic settings as well for the military.
Medical College of Georgia neuroscientist Dr Jay Hegdé and his colleagues developed a relatively rapid method for training civilian novices to become expert camouflage breakers, a skill that even allowed them to sense that something was amiss even when there was no specific target to identify.
Experienced radiologists also have this intuitive sense, detecting subtle changes in mammograms, sometimes years before there is a detectable lesion. One of the main goals of radiology education is training novices to develop advanced or ‘expert’ search methods to improve their recognition of abnormalities, While artificial intelligence may significantly improve diagnosis, there is also the potential to improve the skills of humans.
The researchers behind the camouflage breaking technique wanted to know if trainees could detect the actual camouflaged target or just sense that something is out of place, an issue that is highly significant in real world circumstances.
They already knew that they could train most nonmilitary individuals with good vision to break camouflage in as little as an hour daily for two weeks, which could benefit the military.
“The potential for rapid training of novices in the camouflage-breaking paradigm is very promising as it highlights the potential for application to a wide variety of detection and localisation tasks,” said Dr Frederick Gregory, programme manager, US Army Combat Capabilities Development Command Army Research Laboratory. “Results in experts highlight an opportunity to extend the training to real world visual search and visualisation problems that would be of prime importance for the Army to solve.“
This sort of enhanced ability to spot something amiss could have great applications in medical diagnosis and in search and rescue situations, to name a few.
For this study, six adult volunteers with normal or corrected-to-normal vision were trained to break camouflage using Hegdé’s deep-learning method, but received no specific training on how to pinpoint the target. Participants viewed digitally synthesised camouflage scenes such as foliage or fruit and each scene had a 50-50 chance of containing no target versus a camouflaged target like a human head or a novel, 3D digital image. Similar to how computer scientists ‘trained’ self-driving cars, the idea is to get viewers to get to know the lay of the land that is their focus. “If it turns out there is something that doesn’t belong there, you can tell,” he said.
Trainees could then either look at the image for 50 milliseconds or as long as they wanted, then proceed to the next step where they quickly viewed a random field of pixels, that work like a visual palate cleanser, before acknowledging whether the camouflage image contained a target then using a mouse to show where the target was. “You have to work from memory to say where it was,” he notes.
When the participants could look at the image for as long as they wanted, the reported target location was not much different from when they only had 50 milliseconds — which is not a lot of time for their eyes to move around, Dr Hegdé said.
Again, participants had no subsequent training on identifying precisely where the target was. Yet even without that specific training, they could do both equally well. “This was not a given,” Dr Hegdé noted.
In a second experiment with seven different individuals they used a much-abbreviated training process, which basically ensured participants knew which buttons to push when, using a clearly more pronounced ‘pop-out’ target with scenarios like a black O-shaped target among a crowd of black C shapes. Both the longer and shorter viewing times yielded similar results to the more extensively trained camouflage-breakers both in accuracy and reaction time.
Camouflage is used extensively by the military, from deserts to jungles, with the visual texture changing to blend with the natural environment. “You often are recognised by your outline, and you use these patterns to break up your outline, so the person trying to break your camouflage doesn’t know where you leave off and the background begins,” he said. Animals have also used camouflage for millions of years to evade predators, or to sneak up on prey.
Context is another important factor for recognition, he pointed out, giving the example of not recognising a person whose face you have seen several times when you see them in a different setting. His current Army-funded studies aim to further explore the importance of context, and the ramifications of ‘camouflage breaking’ in identifying medical problems.
He noted that even with his training, some people are inherently better at breaking camouflage than others (he is really bad at it, he admitted) and the reason why is a goal for future research.
Source: Augusta University
