optical illusions in film

The Eyeline Match: An Exploration

Abstract:

This project investigated the Eyeline Match, an optical illusion commonly used in film and television. This was done primarily by constructing raw footage with a Kodak zi8 HD camera, and editing the footage to create experimental clips of eyeline matches. The clips were shown to a participants whose task was to decide whether each eyeline match worked or not. The results demonstrate some of the parameters that determine the effectiveness of the illusion.

Introduction:

Nearly every film released today makes heavy use of the editing technique called the eyeline match (EM). An eyeline match is composed of two shots, one of a character gazing off-screen, and a shot of the person or object that is the target of the gaze. For example, in Alfred Hitchcock's film Rear Window, the main character is confined to an apartment and often looks out of its rear window at events in the building across from him. Hitchcock employs the eyeline match frequently, cutting from the character looking off-screen to the target of his gaze.

EM is a technique that has been in use since the early days of movies, at least as far back as 1929, to aid the viewer in imagining a consistent virtual space. It was first discovered and studied by the Soviet Montage school of filmmakers. One of the famous experiments from that era was the Kuleshov Experiment, conducted in 1929, where shots of an actor looking straight ahead impassively were inter-cut with various meaningful images (a casket, a bowl of soup, a woman, and so on). Viewers "raved about the acting.... the heavy pensiveness of his mood over the forgotten soup, were touched and moved by the deep sorrow with which he looked on the dead child, and noted the lust with which he observed the woman. But [the experimenters] knew that in all three cases the face was exactly the same." The experiment demonstrated that editing can change a viewer's interpretation of images. (Source: Pudovkin, "Naturshchik vmesto aktera", in Sobranie sochinenii, volume I, Moscow: 1974, p.184. )

Filmmakers have since then taken the eyeline match for granted as part of their arsenal of editing techniques. And while there has been some experimentation (for exampe, by influential filmmakers Jean-Luc Godard and Ozu), there has not been much work done in trying to discover what parameters are necessary for EM to work, or even why it works.

I have chosen to study the eyeline match as my optical illusion because it is so prominently used and yet has been studied so little, and also because it's seldom classified as an optical illusion. An optical illusion is defined to be an optical phenomenon that results in a false or misleading visual impression. An eyeline match, then, which is composed of two static shots to give the (possibly misleading) impression of someone in one shot looking at something in the other shot, definitely counts as an optical illusion. So it is more than just a film editing technique. The methods that I used to study this phenomenon are described in more detail below.

Methods:

Raw footage was obtained using a Kodak zi8 commercial HD camera over a period of two weeks. Editing the footage was done in Adobe Premiere. About 200 minutes of footage was acquired in total, condensed into about fifteen 30-second clips. Experimental clips of eyeline matches were created, and filler clips were obtained from DVDs of hollywood movies. The clips were compiled into one 20-minute-long low-quality video file and was shown in shuffled sequence to participants. Their task was to decide whether each eyeline match worked or not, and to write “y” or “n” on a sheet of paper next to the corresponding clip. A total of seventeen people helped by either being a participant, offering suggestions, holding a camera, or being on camera (including Toronto Police Chief Bill Blair).

Results:

The variables investigated are listed below.
They are Followed by images that demonstrate first shot and second shot of each 2-shot sequence. Below the image, is a description of each shot.
Finally, a percentage (%) shows how many people thought the Eyeline Match worked.

Plausibility of Gaze Direction

(Low-angle shot of a person looking down / Slightly high-angle shot of the front of a building)
% of People who thought it worked: 0% (0/10).


(Low-angle shot of a person looking up / Slightly high-angle shot of the front of a building)
60% (6/10).

Order of shots

(Person looking at the top-left corner of offscreen space. / Toronto Police Chief Bill Blair looking at the bottom-right corner of offscreen space.)
% of People who thought it worked: 30% (3/10)


(Toronto Police Chief Bill Blair looking at the bottom-right corner of offscreen space. / Person looking at the top-left corner of offscreen space.)
% of People who thought it worked: 70% (7/10)

Consistency between orientation of eyeline and camera

(Person with his head tilted 45 degrees looking offscreen right. / 45 degree canted shot of a building.)
% of People who thought it worked: 90% (9/10)


(Consistency between orientation of eyeline and camera. Person looking offscreen right. / 45 degree canted shot of a building.)
% of People who thought it worked: 50% (5/10)

Consistency of camera movement

(A person walking looking down. Camera is low-angle, and moving with him. / High-angle shot of the road. Camera is moving.)
% of People who thought it worked: 100% (10/10)


(A person standing still, looking down. Camera is low-angle and not moving. / High-angle shot of the road. Camera is moving.)
% of People who thought it worked: 60% (6/10)

Similarity of environment

(A person looking right. It is ambiguous whether whether he is filmed at night or day. / Shot of a building exit. Daylight streams in through the glass.)
% of People who thought it worked: 90% (9/10)


(A person looking right. It is ambiguous whether he is filmed at night or day. / Shot of a building exit. We can see it is night through the glass.)
% of People who thought it worked: 100% (10/10)

Sound

(Toronto Police Chief Bill Blair looking at the bottom-right corner of offscreen space. / Person looking at the top-left corner of offscreen space.)
% of People who thought it worked: 77% (7/9)


(with sound)
(Toronto Police Chief Bill Blair looking at the bottom-right corner of offscreen space. / Person looking at the top-left corner of offscreen space.)
% of People who thought it worked: 100% (10/10)

Number of cuts

(Toronto Police Chief Bill Blair looking at the bottom-right corner of offscreen space. / Person looking at the top-left corner of offscreen space.)
% of People who thought it worked: 77% (7/9)


(Toronto Police Chief Bill Blair looking at the bottom-right corner of offscreen space. / Person looking at the top-left corner of offscreen space.)
% of People who thought it worked: 70% (7/10)


The following are not controlled variables, but interesting results nonetheless:

Does EM work for virtual characters?

(Animated girl looking at the bottom-right corner of offscreen space. / Animated skeleton man looking at the top-left corner offscreen space.)
% of People who thought it worked: 100% (10/10)

Does EM work for virtual characters?

(Animated army general looking straight at the camera. / Animated girl looking at the bottom-right corner of offscreen space.)
% of People who thought it worked: 100% (10/10)

Can constructed space be re-constructed when new information is available? (part 1)

(Person in jacket clearly at the left wall of room looks offscreen right. / Girl looks offscreen left.)
% of People who thought it worked: 90% (9/10)

Can constructed space be re-constructed when new information is available? (part 2)

(Girl looks offscreen left. / Person in long shirt looks offscreen to the bottom-right)
% of People who thought it worked: 80% (8/10)



Discussion:

There is a slight anomaly in the data, in that clip 18 was played without sound for only 9 people due to a slight oversight. However, because percentages are being used, the data is still comparable. We now engage in a short discussion of what the above data tables suggest.

Firstly, of all the variables, manipulating the plausibility of gaze direction (clip 4 versus clip 10) elicited the most noticeable reaction. Thus it seems that when the direction of the person's gaze does not make sense with the subsequent shot, the eyeline match does not work for most people.

Order of shots was the next most volatile. In clip 12, shot A was Bill Blair looking offscreen bottom-right and shot B was someone looking offscreen top-left, and 70% of participants said this eyeline match worked. Conversely, only 30% said it worked when the shots were reversed. This shows that the order of shots does matter, though it is unclear why.

When there was consistency between the orientation of the person's eyeline and the angle of the camera (for example, in clip 7 where they are both angled at 45 degrees), the eyeline match worked for 90% of people, as opposed to 50% in the control where the person's eyeline was not tilted. Thus, clearly, eye orientation and the orientation of the camera in the match shot is an important factor.

When camera movement was consistent across shots, people were much more likely to feel that the eyeline match worked as opposed to when the first shot had a static camera, and the second had a moving camera. This makes sense if the viewer considers shot B to be from the point of view of the person gazing in shot A. Thus, if the person is moving, and the viewer perceives that he sees a moving ground, there is consistency. Conversely, if the person is still and the ground is moving, there is less consistency, and the eyeline match may fail.

Clips 9 and 15 taken together show that if a person is gazing offscreen in an ambiguously lit environment (that looks as though it could be either night or day), the eyeline match works regardless of whether next shot presents a night scene or a day scene. Thus, it seems that viewers are pretty permissive about lighting conditions as long as they are plausible. If the lighting is not plausible, however, the inconsistency would probably make the eyeline match fail (for example, if the person was clearly in a night scene looking offscreen, and the subsequent shot was a day scene). This is something to test in a future study.

When clip 18 was played with the sound from a conversation playing across the two shots, 100% of the participants felt the eyeline match worked in that case, as opposed to 77% without sound. Thus, consistency of sound appears to be a very important factor in the believability of an eyeline match.

Finally, increasing the number of cuts seems to make the EM more convincing as well. Clip 18 cross-cuts between shot A and shot B four times, and this had a slightly higher (77%) number of people who believed the match worked, whereas clip 12 just showed each shot once and had a success rate of 70%. Thus there may be a “reinforcing” that occurs in the viewer when they are witnessing a series of cuts back and forth between two shots.

Besides the controlled variables, there are two other interesting facts to note from the survey.

Firstly, eyeline matches are not restricted to filmed people – they work just as well for animated characters, as all 10 participants felt that the 2 eyeline matches between animated characters worked.

Secondly, it appears once a viewer constructs an imagined space based on information from an EM, they are willing to reconstruct that world to accommodate new information that may be given to them. For example, in clip 2, a guy in a jacket looks to the offscreen right, and there is a shot of a girl. The girl looks to the offscreen left, and there is a shot of a different guy. Yet most participants felt that both eyeline matches worked. This means that after mentally constructing a world in which the guy in the jacket is to the left of the girl, participants were then willing to modify the world to have the second guy placed to the left of the girl. Thus it seems that viewers are permissive about modifying the world via eyeline matches.

Conclusion:

In general, it was somewhat surprising that most of the eyeline matches above were considered by people to “work”. The results show an underlying theme of permissiveness on the part of the viewer: when it comes to evaluating whether or not an eyeline match worked, viewers seem to want to believe that it works unless they have a strong reason not to (for example, if the gaze direction is not plausible).

Another theme was consistency. Where there was greater consistency between shot A and shot B, participants were more likely to feel that the EM worked. This was seen all over the place with respect to gaze direction, orientation, camera movement, environment, and sound.

If there were an opportunity for a future iteration of this experiment, some things I would improve include having a larger sample size, splitting the sample into multiple groups (for example, to show control and non-control versions of clips to different groups so that memory would not be a factor), and having multiple clip varieties per variable for redundancy.












old stuff:

Week 9:

The Eyeline Match: An Exploration

Week 8:

Progress update:EM

Week 7:

reading week! just some slight refinements.

Week 6:

Refined sketch with slightly clearer goals: here.

Week 5:

Week 4:

Project Sketch:

Sketch slideshow.

Week 3:

Illusion attempts:

1. My Reverse-Phi Illusion Attempt

Failed attempt. (frame rate too slow)
(Original)
Explanation.
Another Variation.

Project:

Illusions in Films. A lot of films make use of optical illusions and visual phenomena. Study these and maybe devise one. Two of the more dramatic examples:

-Forced Perspective
Employs optical illusion to make an object appear farther away, closer, larger or smaller than it actually is. (Wiki) -Eyeline Match
Creates in the viewer's mind a subjective reality that may differ from the objective reality. The thing the person is supposedly looking at may not be in the same room.

Week 2:

Gestalt Illusions:

Definition: configuration or pattern of elements so unified as a whole that it cannot be described merely as a sum of its parts
(wordnetweb.princeton.edu/perl/webwn/gestalt)

Examples from wikipedia:

Examples from michaelbach.de:

Examples from M.C. Escher:

Escher did numerous works where it's hard to distinguish figure from ground: His works also had numerous "impossible scenarios", variations on impossible figures:

Examples from Dali:


Week 1:

Todo:

  1. Explore illusions on the web. Start with:
    -http://www.michaelbach.de/ot/
    -http://www.optillusions.com/
    -http://en.wikipedia.org/wiki/Optical_illusion
  2. Using the Emerging Images paper as a gold standard, think of ideas for a project.
  3. Collect Gestalt illusions to discuss for next week.