Would using the "flaws" in human perception of an optical illusion graphic allow a system to differentiate between humans and machines (using these pictures like a CAPTCHA)?
Some examples are where a human might count "dots" (that are not really there) or choose which parts of a picture have matching colors (when the colors don't actually match due to the illusion).
All the cars above are the same size, but humans see one as larger.
The orange circles above are the same size, but humans see one as larger.
The squares above are same color, but humans see the bottom one as lighter. Cover the line the arrow is pointing at with your finger and you will now see the squares as a computer would.
If a computer were asked to pick out the largest car or circle or the lighter square from these pictures, it would have more difficulty than a human would.