
"From the Rorschach inkblots to the famous ' duck-rabbit', scientists use an arsenal of so-called ambiguous images to help probe the human mind. The team from Johns Hopkins University used AI to create 'visual anagrams'. Each picture contains two animals, but scientists say you don't have a choice over which animal you can see. Just like an anagram spells a different word when the letters are rearranged, these visual anagrams have been designed to show a different image when rotated."
"For example, one visual anagram shows a bear when shown in one orientation and a butterfly when rotated 90 degrees. Lead author Tal Boger, a PhD student studying perception, told Daily Mail: 'Something special about visual anagrams is that you actually don't get much of a chance to see them one way first. 'They let us take the exact same image and make you see it in a different way.'"
"Messy, chaotic information from our eyes is transformed into data that the brain can make sense of through a process involving dozens of alterations, assumptions, and omissions. That means our perceptions of objects out in the world are influenced by many more factors than simply what we are seeing. This is a big problem for scientists trying to study perception, since it's very hard to know which aspect of a picture is having the biggest effect."
Ambiguous images have long been used to probe perception. AI-generated visual anagrams contain two animals within the same image and are designed so rotation produces one animal or the other. One example presents a bear in one orientation and a butterfly when rotated 90 degrees. Visual perception transforms messy retinal input through many alterations, assumptions, and omissions, causing perceptions to depend on factors beyond the raw image. Those many influences complicate experiments because different objects vary in multiple correlated ways. Visual anagrams allow the exact same image to produce different percepts, aiding isolation of perceptual mechanisms.
Read at Mail Online
Unable to calculate read time
Collection
[
|
...
]