Mixing Memory

UPDATE: I’ve messed with some of the images below the fold, which will hopefully make it easier for people to see the illusion without having to move all round the room.

Last year, Rob Jenkins published a seriously spooky-looking illusion (it freaks my son out) in the journal Perception (1). Take a look at this face (from Jenkins’ paper, Figure 1, p. 1266):

Spooky, right? Hopefully you all see a spooky looking woman (it’s actually a combination of two female faces, which is why it looks so creepy) who is looking to your left (her right). Now take a look at the face again, only this time, much smaller:

Now she’s looking to your right! What the hell, right? Try the animated version:

If you doubt these are all the same image, then you can test it for yourself. Look at the big face again, and then take a few steps backwards. You should get the same effect: she starts out looking to your left, and ends up looking to your right.

What’s going on? Well, researchers studying gaze perception had hypothesized that we determined the direction of a person’s gaze using a very simple cue: darkness. The pupil is dark, and in most cases, the iris is dark relative to the sclera (the white part of the eye), so it’s a good bet that whichever side the dark is on, that’s the direction a person is looking. However, Jenkins created the above face by superimposing two faces, one looking to the left (your right), and one looking to the right (your left). He filtered the two differently, so that one’s iris would be dark, and the other light (the dark one is, obviously, on your right).

When he did this, Jenkins had immediately created a problem for the darkness = gaze direction hypothesis, because when you view the face up close, it appears to be looking to the left, despite the fact that the right side is much darker. Why? Because, Jenkins believes, up close, you’re able to barely see the pupil, and see the iris fairly well, which overrides the darkness cue and tells you that the woman is looking to the left. When you get further away, however, the iris’ edges become blurry and blend in with the sclera, and the pupil is very difficult to see, so the darkness cue kicks in again (you get the same effect when you blur the image, again suggesting that it’s the availability of other cues that determines whether the darkness dominates).

This discovery led Jenkins to produce an even more striking version of the illusion, which he calls the “Ghostly Gaze Illusion,” and which won him second prize in this year’s Best Visual Illusion of the Year contest. In this version, close up, two faces appear to be staring at you, while far away, they appear to be looking at each other. You can see this version here. The explanation for this is the same: when the image is big (up close), you can see the outlines of the irises and, in this version, the pupils are very clear. From far away, though, the dark in the corners of the eyes is more salient, causing the faces to appear to be looking at each other.

Jenkins, R. (2007). The lighter side of gaze perception. Perception, 36(8), 1266-1268.

Comments

#1 Optimus Primate
June 23, 2008

Try as I might, I cannot see the woman smaller image as looking to the right of the screen. I see her as cross-eyed.
#2 cephyn
June 23, 2008

When I watch the animated one, I still see her looking to the left. and if i immediately look up at the small image, she’s looking left.
#3 NPD
June 23, 2008

Likewise, to me she’s looking left. After reading the text I went back and could sort of make myself see her looking to the right, but it took effort.
#4 Justin
June 23, 2008

Both my girlfriend and I, looking at the illusion on laptops, had to move the screen a long way away to get the illusion. I was all set to tell you it didn’t work until I put the computer past arms length.
#5 MRW
June 23, 2008

Even the shrunken image is too large for the effect to work on my screen while sitting at a normal distance from the screen, and it sounds like other people are having this problem, too. Take a few steps back from the screen and it works much better.
#6 Tim
June 24, 2008

This is very similar (in fact, almost the same) as an illusion pioneered by Aude Oliva in the 90′s. She makes images that combine two images and look like one image from close up and another from far away:

http://cvcl.mit.edu/hybrid_gallery/gallery.html
#7 Mac
June 24, 2008

I had to roll my chair pretty far back to get any change. Then it looked like she was looking straight at me,, which is freaky.
#8 Luis
June 24, 2008

When I watch the animated one, I still see her looking to the left. and if i immediately look up at the small image, she’s looking left.

Me too. But the first time I looked to the small image I thought it was inverted. The illusion exists but it’s diffuse.

The other illusion is much more clear: I always see the two “twin” women looking to each other until I get like 1/3 close and can pick apart their white iris.The shadows have been manipulated anyhow.
#9 Samantha Vimes
June 24, 2008

I not only saw the illusion, I also realized approximately what had happened after a second look at the second image without scrolling down.

But I am an artist, and we have to train our eyes to see how light and shadow effect what we see.
#10 Chris
June 25, 2008

I see a similar illusion when looking at the Mona Lisa.
#11 Michael
June 25, 2008

Has anyone else noticed that when you look at the image from a “middle distance” the woman appears to be cross-eyed, which negates spookier elements just a bit!
#12 web design company
June 25, 2008

i dont see it.
#13 Tom
June 26, 2008

Try squinting at the large image. The eyes with change direction. At least for me.
#14 Tom
June 26, 2008

Try squinting at the large image. The eyes will change direction. At least for me.
#15 jon
June 26, 2008

Actually, even though the source files of both the large and the small images are the same, when an image is altered in size digitally, the pixel information that is present in the image is altered, or interpolated, it has to degrade, as the resolution available is a fixed value, in this instance 72 dpi (screen resolution).

When the image is reduced enough, what you start to get is an interpretation of the larger image where pixels become merged, the values are evened out, usually in favour of dark and light, the mid tones are ignored, as your eye will blend the two to make up for lack of detail. If you look at the eye in the large image, the Iris and the pupil, the darker tones against the white of the eyeball, have an overlay making them mid-toned, this will change value as the image gets smaller.It is this change that allows the eyes to appear to change the direction that they are looking.

The effect is undoubtedly a natural optical one, our own perception tends to “fill in the gaps” or interpolate information that should be or is expected to be there, but which may not yet be visible. We use this to recognise things at a distance, especially things that are familiar.Human faces being one of the first things as babies that we are “tuned” to recognise. In this instance however, the image software used to generate the picture is also contributing significantly to the effect.

Thus it is not strictly speaking true to say that they are the same image, just one is larger than the other. In actual fact the larger image has much more information than the smaller one, and this information is visible to us.

Interesting though, reminds me of some people in my family!
#16 Selena
October 26, 2008

Woah my eyes hurt.
It is just the darkness
A trick
Happy diwali ppl
#17 Anonymous
July 21, 2010

I see her eyeballs moves to the right side of mine