Saturday, September 19, 2009

Eye Tracking and Composition, Part 2

Below is a scanpath image of the artwork that we saw in yesterday’s post. The chart represents the behavior of an individual who, with no prompting, looked at the artwork for a sixteen second period on a computer screen.

The computer recorded a series of circles, indicating where the eye paused momentarily, connected by a thin blue line.

The scanpath reveals that the eye darts unpredictably in straight jagged leaps known as saccades. Saccades occur between three and five times per second, alternating with brief periods of rest called fixations.

The white glow around each circle represents the subject’s peripheral vision. (The heavier blue shows a running average of the center of attention and the orange line is an attempt by the computer to detect reading behavior. Those lines are not important for the study of artwork.)

The numbered black boxes are time markers, indicating the position of the eye at each passing second. The session begins at the green dot and ends at the red dot, the last point of rest before the image disappeared. By following the blue line second by second, you can precisely reconstruct the viewer’s experience.

The test subject’s eye enters the composition at the top center and zigzags down to the figures at left center. This happens within the first second. In the next three seconds it swoops to the right, leaps upward to glance at the upper right corner, and then moves across the center of the picture in large strokes, pausing briefly to look at the near and far buildings.

For the remaining ten seconds the subject’s gaze slides back and forth in smaller saccades, examining the people in the scene.

According to Greg Edwards, President and CEO of Eyetools, “During the first 3 1/2 seconds, this particular person was getting the lay of the land. How long people take to get this initial overview will depend on each picture. They’re trying to understand the basic structure or the context of the picture.”

After that, they usually settle into finer eye movements. “If they make a big movement,” he said, “they’re typically searching for context. If they make a smaller movement, they’re looking for detail.”

The second person’s scanpath (above) both resembles and differs from the first one. The eye also makes large orienting moves initially, taking in the far vista and the full array of people below. But this scanpath shifts between large and small movements throughout the session and spends more of the time looking at the distant vista and the surrounding architecture.

It might be hard to make out these diagrams in small Web illustrations. For the sake of clarity, this video roughly reconstructs the sequence of saccades over the same approximate overall duration——though it doesn’t accurately represent the relative duration of each fixation.

Tomorrow we’ll see what we can learn from crunching together data from a lot of different observers, and I'll suggest some preliminary conclusions.
Thanks, Greg! Link to Greg Edwards's Eyetools blog. and Eyetools website.

Related posts on GurneyJourney:
Eyetracking and Composition, part 1
Eyetracking and Composition, part 2
Eyetracking and Composition part 3
Introduction to eyetracking, link.
How perception of faces is coded differently, link.


C B Sorge said...

Really neat! I'd love to have this technology available to artists regularly. I'm very curious what conclusions you come to.

I've seen studies done with eye tracking that also used visual artists in the study, and their scan paths were markedly different then laypersons. Artists were more likely to look all over a picture, where as someone else would mainly focus on faces. Unfortunately I can't bring up a link.

Debbie Alden said...

This is a great blog entry, how incredible to know that such technology is out there and am looking forward to future entries on your conclusions, findings, etc.

PS-Great blog, visit often, but first time commenting.

badbot said...

according to these two different test, both look at the column with theses spikes first. I think that i first looked a these too.
maybe it's because it seems the area combining two great contrast, form and values... or is it due to a certain golden rule?

wonder on what conclusion this series of post will end up to!

Erik Bongers said...

It seems to me that the first person was looking at what 'happens' in the picture - a strong focus on the people and other things that have 'meaning' like the texts(even though they are not readable).

The second person seems to more evenly devide his/her attention accross the picture. There's appreciation for lots of purely visual elements, such as the wall and roof on the left and distant castle's roof (two fixations).

I wonder if it's possible to draw conclusions from someones fixation points. Is the second person more into visuals (or art) than the first person?

As an artist, I also appreciate the abundant variation in clouds, mist and haze in this picture, yet non of the two test subjects pauzed at those areas. So can I bluntly conclude that none of them were artists themselves or is all of this pure nonsense?

(I like to divide people into visual and non-visual people and these experiments might hint if there's some truth in that or not)

Steve said...

Fascinating topic. I wonder what would happen if you took the tracking lines of 100 people and overlaid them. Would one primary pathway through the artwork emerge?

James Gurney said...

Erik and CB: You've raised a question that intrigues me, too, whether artists look at images differently than non-artists. Unfortunately in this batch of data, I don't have subject profiles or followup interviews.

In any event, it's very difficult to infer too much about the personality of a particular viewer or even the viewer's state of mind from particular fixation points or even from scanpath pattern.

I would guess that your suggestion is right, though, Erik, that the first subject was more interested in the people, and that the second subject was more interested in some aspect of the overall structure.

Badbot: I should have explained that the entry and exit points in these tests is not too meaningful. It's the point that the subject's eyes happened to be resting on when the computer screen changed from the previous picture to this one. The subjects were presented with a series of different images and looked at them for a fixed interval before the image moved on.

Deb, you made my day with your kind comments. Thanks for visiting the blog, and I'm so glad it's helpful.

James Gurney said...

Steve, sorry I missed your question--we posted at the same instant! The answer is YES. We can get cumulative data of which areas got the most views as well as the statistical number-crunching on which areas tended to be seen first, second, and third, and so on.

But a "primary pathway" based on averaged data would probably be a very generalized, just because all the individual scanpaths are so different. It would be like drawing a cumulative route map of how most American tourists visit Italy.

Paolo Rivera said...

Jim, thanks so much for doing this. I must admit that I've always found superimposed compositional diagrams kind of hokey (at best), so it's nice to see people's actual eye path, as opposed to what some artist or historian think is ideal.

In my experience, I look at faces and high contrast first, so I build most of my compositions around those anchors. As a comic book artist, it's my job to make legible and organized pictures, but I never assume that my reader will look at one thing before another (in a single panel, at least).

The only exceptions are word balloons, which I try to plan for in a left-right-top-bottom order, but that has everything to do with reading strictures and little to do with natural eye paths.

Katherine Kean said...

I find this really fascinating. As others have, I also wonder how many, if any of the viewers are artists and also male or female, young or old, etc. I wish I could use one of these myself! I look forward to the conclusions.

Nasan Hardcastle said...

I had no idea how detailed the tracking details would be on this experiment.
After reading you last post (part 1), I was suspicious to learn whether the eye really would follow a pattern reinforcing such things as the golden section, rule of thirds, etc . . .
How interesting, also, to learn that the viewers eye work from general context to specific detail when looking at a scene.

Steve said...

After reading this in the morning, we spent the day at the Boston Museum of Fine Arts and the Gardner Museum. While contemplating the many paintings by Sargent, Rembrandt, et al I felt hyper-aware of how my eye moved over the masterpieces, visualizing my "scanpath," and wondering what anyone else would make of it!

Stephen James. said...

Very ironic that you post about this now, we've been talking about this in class over the last few weeks.

Roberto said...

Deb- I checked out your blog, thanks for the links to Bernie Fuchs and the U-tube. Apatoff’s site is very cool.

As usual, Erik’s analysis is very good, but I think it would be difficult to draw any conclusions as to personalities or professions from eye scans. (I like to divide people into two categories also: those who divide people into two categories, and those who don’t ;)

I am very surprised by how little the scans seem to follow the general rules of composition once they entered the picture at the point of highest contrast and compositional dominance –the column at the golden mean. Yet you say that this entry point is simply the result of random chance, an artifact from the previous image! Vvvvery interesting. Yet in spite of all the apparent chaotic zig-zagging and anti-compositional cohesiveness of both scans, both subjects ‘read’ the paintings main elements, the horizontal foreground structures and figures, the background castle, and even the sign in the upper right corner.
I’m thinking that quite possibly the aesthetics of pictorial composition and design may function on a more holistic-psycho-intuitional-cultural level, while the eye movements have more to do with a primordial-evolutionary-visual-input-reflex-system-thing. In other words: how we see or gather information is different than, or independent from, what we perceive about that info. (I could be totally wrong, but I got to make up some fun words!) -RQ

Anonymous said...

This is a great tool to use in studying compositional techniques. Thank you for posting! I've also always wondered how animals would respond to these eye-movement experiments...

Susan Roux said...

This is very interesting. I wonder how different the eye path would be for an artistic person versus one who is not?

James Gurney said...

Susan, I tackled your question about whether artists see differently in another post:

In a nutshell, YES!