A Model for Controlling the Stimulus Information for the Perception of the Human Gaze Line

August 1967

A Model for Controlling the Stimulus Information for the
Perception of the Human Gaze Line

J. J. Gibson and Stephen West, Cornell University

The World Wide Web distribution of James Gibson’s “Purple Perils” is for scholarly use with the understanding that Gibson did not intend them for publication. References to these essays must cite them explicitly as unpublished manuscripts. Copies may be circulated if this statement is included on each copy.

This paper discusses the construction of a model of the human face in which eye movements can be accurately controlled by the experimenter, so that any bias of using a human “looker” in gaze experiments can be eliminated.

A life mask was made of a preparation of Epoxy resin and cut-up fiberglass, which was shaped over a rubber mold of an actual human face. A primary mask was made by applying petroleum jelly to a human face, (with suitable protection for the model ‘s eyes, nose, and mouth) and then painting “Pliatex” mold rubber over the entire face. The substance was applied in several layers, allowing each layer to dry before beginning the next. After the “negative” rubber mask had been completed and removed, the Epoxy preparation (resin and fiberglass) was painted on its inside surface. The resin and fiberglass was built up layer by layer, sanding off the rough spots after each application. This completed mask was given a final sanding (after having been removed from the rubber mold) and was painted using a beige spray paint which seemed to closely approximate fleshtone. After this work was completed and the eye sockets were formed (see below), the mask was mounted on the front of a 15″ x 24″ plywood panel in which a hole had been cut to match the mask. The mask was about 10″ high by 5″ wide.

The blown-glass eyes were obtained from a supply house for artificial eyes (Mager and Gougelman, 30 N. Michigan Ave., Chicago, Ill.). The pair of eyes used in this model has blue irises and are matched as closely as possible.

The eye globes are coated with petroleum jelly in order to allow free movement in the sockets. The sockets were made of “Pliatex” mold rubber and were painted on layer by layer over the lubricated eyes (none was painted on either the iris or the front section of the sclera). The sockets were then fitted into the model and fastened with contact cement.

When not in use the eyes are kept packed in tissue paper, as they break very easily when dropped. The eyes are mounted as follows: a small part of the rear section of the glass globe was drilled out so that a 1/8″ diameter straight steel drill rod could barely pass through. This drill rod was carefully fastened to the back of the iris with Epoxy glue. Before the glue set the rods were rotated and the rod thus aligned closely with the axis of the eye.

The drill rod sections are about 16″ long. The protractor is 7.0″ in radius. The protractor is placed directly in back of the model’s left eye, 14″ from the pivot of the eye. The amount of turning of the eye will then be one-half of the angle measured by the protractor (theorem from geometry). The zero reading or calibration of the scale was determined by having a number of observers judged when the model was looking at them. A variable linkage rod between the two log rods is provided to permit the experimenter to vary the convergence of the eyes. The linkage rod is fastened by a pin to each of the long rods to allow for freedom of movement of the eyes. In the present model the eyes cannot be moved up or down although the model could be modified to do so.

E operates the eyes i.e. points the gaze lines by moving the rod until the pointer is on the desired degree-reading. Between trials a cloth screen is thrown over the face to prevent the Ss from watching the movement of the eyes. The Ss will sit on a stool of variable height directly in front of the model. The stool will be adjusted so that the eye level of the subject is in the same horizontal plane as the eye level of the model. The convergence of the eyes can be varied by adjusting the turnbuckle in the variable linkage rods. Strong frontal illumination will be used in the experiment.

Notes for experimental investigation, based on Gibson and Pick, 1963, and Cline, 1967 (both Amer. J. Psychol.)

1. What does the subject perceived when the eyes are focused at infinity? How does this compare with when the eyes are focused at 6 feet? Predict: Convergence at the proper distance is necessary fo S to perceive he is being looked at. Perhaps a second study could be done in which the amount of convergence is varied, thus determining the limits within which this variable can be perceived.

2. Cover one eye with a patch. Does the position of one eye convey enough information for the observer to perceive he is being looked at? Predict: both eyes are very necessary for the perception of being looked at (i.e. great increase in acuity).

3. Our preliminary investigation showed clearly that the centering of the iris within the oval frame of the eyelids is the important information that is utilized in the perception of gaze direction. When the eyes were pointing straight ahead, but were not centered in the oval frames, several observers agreed that the eyes were not “looking at them,” i.e. were “crossed.” However, when the eyes were moved so that they were centered in the oval frame, but were not pointing ahead, the observers agreed within close limits that they were being “looked at.” The relation of the irises and sclera to the eye frame is then the essential information with which we are concerned.

The available optical information isolated and controlled by this model is the centering of the iris relative to the oval frame of the eyelids (i. e. the sclera). With the head turned the other feature of the face-from, especially the nose, become important in determining the direction in which the model is looking. However since the model has fewer detailed features that a human face, we might expect the background information to be poorer and consequently the judgements of the direction of the model’s looking to be less accurate. However, information from the panel such as the grain of the wood and the edges may compensate for the loss of information avaialbe from the face itself.

With a real human looker we know that the optical information picked up is only contained in the top one-half of the face, at least when the looker has his head straight (c.f. Cline, 1967). With the face straight ahead, the following features should be important in the perception of looking: 1) the convergence of the eyes, 2) the centering or off-centering of the irises in the eye-frames. With the head turned the background features of the face should become very important (especially the nose) in the judgement of gaze direction. The human face provides an enormously detailed background against which this judgement can be made. Also the deformation of the eyelids with the turning of the eyes may provide additional information. Finally, at close distances the fine structure of the eyes (i.e. the texture of the iris and curvature of the cornea may be important. In the prelimarly investigation, it was found that at a distance of the two feet or closer the direction of the eyes was judged independently of the centering of the iris in the oval frame. As the subject moves close to the model, he utilizes information from the eye itself and not from the relationship of the eye to the oval frame in judging the direction of the gaze line.




Footnote: Our thanks are acknowledged to Marc Glassman, Yale University, who designed the original prototype of the looking model and aided in the construction of the present model.