How do we decide what we're seeing?

One of the most famous perceptual demos is the ambiguous image or "bi-stable figure" of a duck - rabbit:

(source: curiouser.co.uk)

As presented, it looks like a duck, but rotated to the right, it suddenly "transforms" into a rabbit. There are also images that can transform simply based on how you look at them. Artists like Eshcher were fascinated by such images, but it was Salvador Dali who took the phenomenon to the next level with paintings like Slave Market with the Disappearing Bust of Voltaire:

(source: Artchive)

It's worth checking out the image on the Artchive site because you can zoom in and look at it in detail. Notice that the two nuns in the center of the picture alternately "disappear" and "reappear" when the larger bust of Voltaire resolves into view.

Why do we sometimes see the nuns and sometimes see the bust of Voltaire? Lizann Bonnar, Frédéric Gosselen, and Phillippe G Schyns of the University of Glasgow devised two experiments to answer that question ("Understanding Dali's Slave Market with the Disappearing Bust of Voltaire: A Case Study in the Scale Information Driving Perception," Perception, 2002 [PDF Link]).

In the first experiment, they showed participants a greyscale version of the painting, cropped in on the "bust," and asked them whether they saw primarily nuns or Voltaire. Then they used a systematic "bubble" filter to enhance or obscure certain features of the painting. They showed dozens of different types of obscured paintings, asking each time whether observers saw the nuns or the philosopher. What they found is that when fine-grained details of the painting were obscured, observers were more likely to see Voltaire, but when the larger-scale features were obscured, observers were more likely to see the nuns. Not too surprising. More importantly, however, from this information they were able to generate a "hybrid" picture, which combined the most potent details signifying "nuns" and the details that led to the "Voltaire" response.

The second experiment is where they struck gold. Relying on research in spatial frequency which suggested that viewers could be conditioned to filter out certain patterns of their visual fields, Bonnar and her colleages showed a new group of participants one of two types of random noise patterns. The first group was shown a moving fine-grained pattern, somewhat like static on a TV screen, for several minutes. The second group was shown a similar pattern, but with larger patches of light and dark. Then each group was shown the hybrid nun-Voltaire image. All of the people who were shown the large-grain pattern said they saw nuns, and 4 of the 5 people who were shown the fine-grain pattern said they saw Voltaire.

So part of why we see one image or another depends on what we saw immediately before. Our visual system filters out the "noise" of the random pattern, and so can only see details that are sufficiently different from the pattern we just saw.