Saturday, May 31, 2008

The peripheral escalator illusion

Here is an illusion that I developed as part of a collaboration with Zhong-Lin Lu (USC) and Emily Knight (then, an undergraduate research assistant in my laboratory at Bucknell). Emily Knight presented this illusion at the 2008 Best illusion of the year contest and as a talk at the Vision Sciences Society 2008 annual meeting (abstract).

You are looking at columns of zebra-like ovals that swing back and forth in front of a diagonally striped background.

1. Look directly at the ovals. The ovals appear to swing horizontally.

2. Focus your gaze several inches above the screen, but pay attention to the ovals as you do so. The ovals should now appear to swing diagonally.

The difference between the two conditions is dramatic. If you don’t see it, try fixing your gaze a few inches higher above the screen.

Comments: If you’ve seen a diagram of a human eye, you know that in the back of the eye is a layer of cells called the retina, and in the center of the retina is an area called the fovea. The retinal area outside the fovea is referred to as the periphery. When you look directly at the ovals in the above display, the image of those ovals falls on the fovea. When you focus your gaze a few inches above your computer monitor, the image of the ovals falls in the visual periphery.

The machinery for foveal vision differs from the machinery for peripheral vision in numerous ways—too many ways, in fact, to list them all in a blog entry. As a result of these differences, we are able to see objects in much more detail when we view them directly (foveally) than when we view them indirectly (peripherally). Our representation of the world, therefore, is much different from the representation in a typical photograph, in which all objects in the same visual plane have the same amount of detail.

The different perceptions of the “peripheral escalator” illusion are partly due to the inability of peripheral vision to see fine detail. You can simulate this aspect of peripheral vision by removing fine detail from the display and then viewing the display foveally. Here are three simple ways to remove fine detail from the display: 1) view the display from about 6 feet away from the monitor; 2) scrunch up your eyes; or 3) if you can't (or can barely) see the words on the computer screen without corrective lenses, take them off and then view the display.

When you remove the fine detail, you will see a weird pattern of stripes that move up and down. This pattern is similar to what people see when they view the display indirectly, but it does not capture the peripheral perception of a diagonal sweeping motion that observers frequently report.

It seems, then, that relative to foveal vision, peripheral vision is missing not only the ability to represent fine detail but also “something else.” The nature of the “something else” is an important question for vision research. One possibility is that peripheral vision is not very good at alignment (in the trade, we say that the visual periphery is poor at representing “phase” information). Indeed, I created the peripheral escalator illusion as part of a search for the perceptual ramifications of the “poor-phase” hypothesis. While the poor-phase hypothesis may be correct, my colleagues and I suspect that central and peripheral vision may also differ in their capacity to segregate individual features. We have designated this possible capacity limitation “feature blur”… stay tuned.

As a final thought, primates differ from other mammals in the way cell types are distributed in the retina (see Masland, 2001, third paragraph). The primate fovea is dominated numerically by a particular group of post-receptoral cells, whereas in the retina of other mammals and in the primate periphery, cell types are distributed more evenly. It may be that most non-primate mammals see the world in a way that is closer to our peripheral vision than to our foveal vision. If this is so, then perhaps lions looking at a pack of moving zebras see something not unlike what we see when we look indirectly at the peripheral escalator illusion.

I will be on the road next week, so I may not be able to post a new illusion until after I return.


Note 6/3/08 8:25: For your reference, here is a link to a pdf of an excellent earlier study (1992) on “Misdirected visual motion in the peripheral visual field,” by R. Cormack, R. Blake, and E. Hiris (Vision Research, 32, 73-80).

As you will see, the principle is the same, but I think there may be interesting differences.

Here is a link to Randolph Blake’s website at Vanderbilt University. I encourage you to view the visual phenomena on his demonstration page.

Saturday, May 24, 2008

Squaring the Diamond

Look at this image at your usual distance from your computer, and then from across the room. Does the image look different when you see it close up vs. far away? (If you don’t feel like getting out of your chair, you can get the same effect by scrunching up your eyes or, for those who wear glasses, by removing them.)

You should get two different impressions of the image. When you are close and the image is in focus, you should see horizontally striped diamonds on a field of blurry vertically striped diamonds. When you are far away (or if you have scrunched up your eyes or removed your glasses), you should see a field of squares.

Why does this happen? We often take it for granted that we see the world at different scales. Look at the page that you are reading: you can probably perceive fine details (say, letters and words), broader features (like the shape of the paragraph), and large features (the outline of the page).

The amount of information at each of these scales changes as we move about the world. When you are far away from your computer monitor, you can’t make out letters or words because they are too small to see. But as you move closer, the letters and words become bigger, and (voila!) this new information becomes part of your perceptual world.

Our ability to adjust to these changes in size is helped considerably because, in the real world, objects tend to scale together: zoom out, you see the page; zoom in and you see words on the page. As you zoom, a word on the page gets bigger or smaller at the same rate as the other words.

“Squaring the Diamond” is compelling (at least to me) because it seems to violate this basic assumption about how objects scale in the physical world.

The display is composed of blurry diamonds and regular (a/k/a non-blurry) diamonds. When you zoom out, the diamonds combine to form a field of squares; but when you zoom in, you can see detail only in the diamonds that have not been blurred. Unlike the non-blurry diamonds and the words on this page, the blurry diamonds do not gain more detail as you move toward them because the blurry diamonds do not contain fine detail (in vision science circles, “fine detail” is known as high spatial frequency information). The result is that your perception of the field of squares breaks apart when you are near enough to see the fine detail in the non-blurry diamonds.

The buttons in the display allow you to blur and unblur the sets of diamonds. Click on the buttons, and see what you can find out about the image. Notice that you see squares when the horizontal and vertical diamonds are both blurred, and when both are unblurred.

The principle is similar to Schyns and Oliva’s hybrid images (Aude Oliva’s gallery of hybrid images is really worth seeing; here is a link to an article about one of their hybrid images: Dr. Angry and Mr. Smile).

Monday, May 19, 2008

Grouping by contrast

Here is an illusion from a brief article that I wrote with Kai Hamburger (“Grouping by contrast—figure-ground segregation is not necessarily fundamental,” Perception, 2007).

The five disks in the top row of the display all change gradually from black to gray to white and then abruptly back to black; in the bottom row, the five disks all change gradually from white to gray to black and then abruptly back to white.

Click on the “add/remove gradient” button to place gradient rectangles behind the disks. The switch in background from solid to gradient creates a dramatic change in the way the disks are perceived. The disks themselves have not changed, but now motion appears to sweep from right to left across each row. The perceived motion is similar – in some respects – to the perceived motion in the window shade illusion, since it follows the location of minimum contrast between the disks and the background.

There are many features about this shift in perception that we could talk about, but I would like to draw attention to what I think the display says about how we organize visual information. Notice that when the solid background is present, you tend to group the disks into two horizontal rows of five, but when the gradient backgrounds are present, you tend to track the motion and group the disks into five vertical columns of two disks each (so far, all viewers have automatically grouped the disks in these ways).

This shift in perceptual grouping is really unusual. When you group the disks into columns, you pair a white disk in one row with a black disk in the other row even though each of these disks has the same luminance level as the four other disks in its row. A horizontal grouping would seem like a reasonable way to organize the disks because – just like in the contrast asynchrony – the disks in each row become light and dark at the same time. However, it seems that the visual system prefers to group vertically because the disks in each column have similar contrast levels relative to their respective backgrounds.

In the article that I link to above, Kai Hamburger and I suggest that contrast-based grouping poses a bit of a puzzle. The Gestalt approach to visual perception proposes that the visual system organizes the world into simple perceptual units in accordance with well-known Gestalt laws (for example, similarity, symmetry, proximity, closure, common fate). Central to the Gestalt approach is the idea that the visual system organizes the world in terms of “Figure” and “Ground” (for an example, see Rubin’s face/vase illusion). Contrast information, however, cannot really be considered part of the Figure perceptual unit or the Ground perceptual unit because the contrast information cuts across the figure/ground border (the contrast information represents the luminance levels of the disks relative to the luminance levels of the background).

The illusion in this post illustrates a condition in which the visual system privileges contrast information over object similarity in order to organize the visual scene. At some level, this should not be surprising; a visual scene can be described in terms of a variety of stimulus dimensions (spatial scale, luminance, contrast, temporal changes, chromaticity, etc.). The visual system contains parallel neural channels, each of which responds to only a small range in a few of these dimensions. Presumably, the neural processes that organize the visual scene must do so by selecting a sub-population of the neural channels. However, since we generally think about the world in terms of objects that remain relatively stable regardless of context, it can be surprising to see the effect that contrast information can have on how we organize the visual scene.


I was on the road last week, attending the Vision Sciences Society conference (sorry for the gap in the blog). In my last post, I mentioned the then upcoming, now past, best illusion of the year contest. Incredibly interesting new illusions were presented, and the event was a great success. My lab had two entries in the top ten, but (alas) our entries did not place 1st, 2nd, or 3rd this year. You can see the winning entries at this link. I also encourage you to visit the website of the major sponsor for the event: the Mind Science Foundation. They have many resources regarding consciousness that may be of interest to readers of this blog.

Next week: an illusion of spatial scale.

Monday, May 5, 2008

Perpetual collision illusion

Here is a sneak preview of one the illusions that my laboratory will be presenting at the Neural Correlate Society’s “Best illusion of the year” contest that will be held next week (May 11th ) in Naples, Florida. The illusion was devised to investigate questions related to contrast and visual grouping (see article).

Warning: the illusion involves rotating, high-contrast diamonds. If you are prone to migraines or epilepsy, or get motion sickness, please do not stare at this illusion.

[The illusion is a flash file and will not appear in an RSS feed]

Description: You are looking at columns of pink and yellow diamonds separated by columns of spinning black/white/gray diamonds. The pink diamonds appear to move to the right; the yellow diamonds appear to move to the left.

There are two main things to notice about the display:

1. The pink and yellow columns are not really moving. Don’t believe me? Click and drag the spinning black/white/gray diamonds to move them out the way. When you do, you will see that the spinning diamonds are placed on top of a completely stationary colored background.

2. The motion is perpetual. The pink and yellow fields seem always to be headed towards (or away from) each other, but they never meet (and they never grow farther apart). This aspect of the effect can be quite mesmerizing, so be careful.

The motion originates from the edges between the spinning diamonds and the colored fields. The edges of the diamonds are tilted at -45 or 45 deg; the motion, therefore, should always shift in an oblique direction. To get a better handle on this, click the “add/remove diagonal bars” button. The diagonal bars cover up opposite sides of the rotating diamonds so that only every other edge is shown. When the diagonal bars are present, the pink and yellow fields move diagonally.

Why, then, should the pink and yellow fields appear to move horizontally when the diagonal bars are not present? Not to be too technical, but it seems to me that either the visual system is computing motion for the colored diamonds from a vector sum of the motion at the edges; or the visual system is using the information at the edges to define an object (in this case, a diamond), and motion for the object takes precedence over the motion that originates at the edge.

I have also included a button that allows you to “add/remove horizontal bars.” The horizontal bars stretch across the image so that the colored diamonds turn into colored triangles. Nonetheless, instead of seeing individual triangular segments, you perceive the image as a series of colored diamonds that appear to move behind a bar. It is as if the visual system joins the triangles to form the diamonds, so that you perceive a "whole" object.

I will be on the road next week (traveling to the conference). I will try to post a new illusion on Tuesday the 13th or Wednesday the 14th.

/*********google analytics*/ /*********google analytics end */