West, Roger W.*
Gaze between individuals is a form of social communication.1,2 We become ill at ease in social situations in which it is difficult to determine where a person is gazing, probably because we are denied that aspect of communication. For instance, we may become uneasy when a person is wearing sunglasses that hide their eyes. Also, if information about the person's emotional state or interest in a conversation is denied us by their refusal to make eye contact, we may make inferences about their confidence, sincerity, or honesty. We may also be perplexed when an individual has a strabismus3 or is wearing a patch, both of which leave only one eye available for gaze assessment. In addition, if the gazer has a lid shape that we are not used to viewing, such as due to trauma or the presence of epicanthal folds, gaze may be hard to interpret.4 Also, it is hard to interpret eye contact from a turned head, which may be one reason why over-the-shoulder gaze may be disquieting.5 For all the above reasons, the study of the perceived direction of gaze and how it is used in different social contexts is important.
Throughout history, some societies have treated brown-eyed and blue-eyed people differently. For instance, the ideal face in Nazi Germany was blue-eyed, and experiments have been conducted dealing with eye color and prejudice.6,7 Moreover, many movie stars are known for the pale blue color of their eyes (e.g., Paul Newman). However, it may be more than the blue color of their irises per se that makes some eyes so intriguing. A light blue iris (and even a light brown iris) makes the pupil much more visible, which may make the interpretation of the eye's direction of gaze different from that of an eye with a darker iris. This can also give eyes with pale irises an interesting appearance. An example is the cover picture that National Geographic selected as its picture of the century. In spite of all the exotic people and scenery that has appeared on its covers, the editors selected the face of a girl who has an extremely captivating appearance, which is obviously because of her pale-eyed stare.8
To date, most studies have not specified the eye color that they have used to study the perceived direction of gaze. However, there is a plausible argument for how gaze from pale eyes (most often blue) might be interpreted differently than gaze from dark brown eyes. In brown eyes, a major cue to the detection of gaze direction is the location of the iris within the lid aperture. Eyes that have light irises have less contrast at the iris border, which may offer a weaker stimulus for gaze detection, but they also have a highly visible pupil that presents a high contrast feature, which may be used as an additional cue for gaze direction. Because the average pupil is decentered 0.25 to 0.50 mm nasally within the iris,9 there may be a conflict between these two cues. Although the typical pupil decentration may seem small, it makes a difference of 3 to 6° in ocular rotation when judging gaze by the iris alone vs. the pupil alone.10 The corresponding distance in the plane of an observer who is 100 cm from the gazer is 5.25 and 10.5 cm, respectively, which can place the perceived direction of intended eye contact entirely off the observer's face.
The assessment of gaze from a pale-eyed individual is probably based on a combination of cues from both the iris and the pupil, but in eyes that have a dark brown iris the pupil location and, in fact, the pupil itself may be hard to observe. This study tests the effect of visibility of the pupil, and whether it is centered or decentered, on the perceived direction of gaze. A model was recruited who had dark brown irises, and his image was Photoshopped to change the iris color to light blue so that observers could judge gaze from the same eyes when the model's eyes were dark brown vs. light blue, and when they were blue, with centered vs. typically decentered pupils.
This study was approved by the Northeastern State University Institutional Review Board, and all subjects gave written informed consent.
A 30-year-old man served as a model for an imaged head that gazed in different horizontal directions. He had dark brown irises, an uncorrected visual acuity for each eye of 20/20 at 80 cm, and a stereo acuity of at least 40 sec of arc. His pupils were decentered 0.41 mm nasally OD and 0.37 mm nasally OS, which was determined by photographing a ruler in the same plane as the eyes and comparing the ruler and iris/pupil using Photoshop. His horizontal angle kappas, measured using apparatus previously described,3,11 were +1.15° (temporal) OD and −0.69° (nasal) OS. The definition of angle kappa and its importance will be covered in the Discussion.
The model's head was stabilized with the aid of a bite bar covered with dental compound that conformed to his front teeth. His head was then photographed with a 6.1 MP Nikon D70 camera while pointed directly at the lens from a distance of 80 cm as he gazed successively at −9, −6, −3, 0, +3, +6, and +9 cm from the center of a horizontally oriented meter stick located within the plane of the camera lens. Images for each of these locations were taken twice so that for each eye, the above positions on the meter stick could be offset 1.61 cm (1.15°) for the model's right eye and 0.96 cm (0.69°) for the model's left eye, to simulate 0-angle kappa for each eye. Additional pictures were taken of his head with his eyes closed and then without the bite bar.
PhotoShop was used to edit the images. An image taken without the bite bar was used to remove the bite bar from one of the images, which then served as a background head on which the eyes gazing in the different directions were layered. Also, the corneal reflection from the light source was removed from each image by copying the surrounding pupil and iris over the reflection. This resulted in a series of head images in which the eyes had natural dark brown irises and 0-angle kappas, which gazed −9, −6, −3, 0, +3, +6, and +9 cm from straight ahead (minus to the observer's left and plus to the observer's right).
An outline of the brown iris OD was isolated as a separate layer, which was then filled with a brown iris color and dark pupil to closely match that of the natural iris. This simulated iris was then layered over the natural iris OD and also flipped horizontally and layered over the natural iris OS, a process which was repeated for each gaze direction. The layered iris was then turned light blue, and both non-flipped and flipped images were layered to replace the brown irises for each gaze direction. These blue irises were constructed without pupils, and also with pupils that were either centered or decentered 0.41 mm nasally. No attempt was made to simulate the darkening of the peripheral iris that is seen in some eyes.12 Each binocular image was duplicated with either the OD or OS lid closed to form a set of monocular images. This gave 70 monocular images and 40 binocular images that were presented to each observer (the 0.0 cm binocular images were viewed twice). Fig. 1 shows the isolated eyes for each of these configurations for binocular straight gaze.
Twenty-five NSU optometry students aged between 21 and 26 years served as observers. All had a best-corrected visual acuity of 20/20 or better in each eye at 80 cm and at least 40 sec of stereo acuity at near.
The observers, run individually, positioned their heads 80 cm from an LCD monitor (1280 × 1024 resolution) and looked over the same horizontally oriented meter stick that was used to guide the model's gaze. Their head location was controlled by resting their noses against a notch in the meter stick at the previous location of the camera lens. As the images were displayed full-size on the screen, the observers reached an index finger under the meter stick and touched its far side to indicate where they judged each image was looking. They were instructed to use their right hand if the gaze appeared to be to the right of the notch and their left hand if it appeared to be to the left. The distances between the point of touch and the notch were recorded to the nearest 0.5 cm.
The images were shown in randomized order within each of two blocks. The first block presented the images with one or the other eye open (monocular gaze), whereas the second block presented the images with both eyes open (binocular gaze). Binocular gaze was shown last so that it would not bias the judgment of monocular gaze. (Straight binocular gaze can be detected by the symmetry in the iris location between the two eyes. The iris location would then be the same for straight monocular gaze.) There was no time limit for each response and the sessions averaged about 40 min.
Fig. 2 plots the judged direction of gaze against the true direction of gaze for the original brown irises. A comparison with Fig. 3A, which was obtained from the eyes with the simulated brown irises, shows minor differences between the two graphs, but the same basic pattern. Some of the differences may be explained by the more abrupt transition at the edge of the simulated irises, which would give a sharper contrast with the sclera. But the differences may be better explained by uncertainty in the placement of the iris layers because, at the 80 cm viewing distance, an error of as little as 0.1 mm in the placement of the simulated iris over the original iris would give a 1.76 cm error in the true direction of gaze. However, any error would be the same for all the simulated eyes. Because the simulated blue irises were modified from the simulated brown irises, they had the same locations, outline, and texture as the simulated brown irises. This allowed the graphs from the simulated brown irises (Fig. 3A) to be used as the standard against, which to compare the graphs from the simulated blue irises (Fig. 3B to D).
Fig. 3A to D shows that, for all the iris types, binocular gaze (OU) was judged fairly accurately when gaze was straight, but the judged direction was increasingly overestimated as gaze was moved to the side. For monocular gaze (OD and OS), for all the iris types except for the decentered pupil (Fig. 3D), the judged direction of the adducting eye underestimated the true direction of gaze and continued to underestimate the true direction as it became the abducting eye past 0.0 cm up until about 3 cm of abduction, at which point it approached the true direction.
The images with blue irises and no pupil (Fig. 1C) were constructed to match the simulated brown irises within which the pupils were poorly visible (Fig. 1B). This tested whether differences in the color or the brightness of the irises alone, without the influence of the pupil, might cause differences in the perceived direction of gaze. The binocular curve for the blue irises without pupils (Fig. 3B) was very similar to the binocular curve for the brown irises (Fig. 3A), but its monocular gaze showed a greater amount of underestimation than the brown irises when the eyes were adducted.
The images with blue irises and centered pupils (Fig. 1D) tested whether a pupil, which adds an additional high contrast cue to gaze, would affect the perceived direction of gaze if its center coincided with the center of the iris. Again, the binocular curves (Fig. 3C) were very similar to those of the brown iris (Fig. 3A), but monocular gaze showed an even greater amount of underestimation when the eyes were adducted.
The images with blue irises and decentered pupils (Fig. 1E) tested the effect that a typical nasal displacement of the pupil from the center of each iris would have on the perceived direction of gaze. Again, the binocular curve stayed about the same, but the monocular curves moved closer to the true direction of gaze and crossed close to the 0 value for true gaze (Fig. 3D). Because their slopes remained about the same as before, the monocular curves closely overlapped each another. It is noteworthy that such a small change in pupil location made such a large change in the perceived direction of monocular gaze.
The curves in Fig. 3 show the relationships between monocular and binocular stimuli within each iris type but do not show very well the relationships between the monocular and binocular stimuli across iris types. Fig. 4 plots OD, OS, and OU as judged-minus-true vs. true and juxtaposes these curves on one graph to better reveal this relationship. These plots directly show the bias from true against an expanded y axis. Because the curves fell very close to one another, analysis of variances were performed across iris conditions for each gaze angle within OD, OS, and OU separately, and, when the analysis of variances were significant at p < 0.05, Tukey post hoc comparisons were made for all pairings of iris types for each gaze angle (Table 1).
Fig. 4 OD, OS shows that the greatest differences in judged gaze were between the centered pupil (BL PC) and the decentered pupil (BL PD) conditions, and Table 1 confirms that most of those differences were statistically significant. For OD, gaze from the decentered pupil was always judged to be to the right of gaze from the centered pupil, whereas for OS, gaze from the decentered pupil was always judged to be to the left of gaze from the centered pupil. In other words, decentration of the pupil caused the judged direction of monocular gaze to shift in the direction in which the pupil was displaced. Comparisons between the other iris types were not as consistently significant.
Fig. 4 shows that for OU exposure, the curves for all the iris conditions essentially overlapped, and the only significant difference being between BL NP and BL PD at −6 cm. For the binocular curves as a group, central gaze was judged fairly accurately, but their slope was greater than that of the monocular curves.
Only a few of the studies that have used live or photographed models to study the perceived direction of gaze have reported the iris characteristics (pale vs. dark, blue vs. brown, pupil centered vs. decentered). Yet, this study shows that the visibility of the pupil and whether it is centered or decentered has a large influence on the perceived direction of monocular gaze. Indeed, this may be the reason for some of the disagreement within the literature.13,14,15 It is urged that future studies report these iris features. Angle kappa, which is another parameter that has been shown to influence the perceived direction of gaze, has also been underreported.10
Cues to the Direction of Gaze
There are two main theories about what features of the eyeball and lids are used to determine the direction of gaze from a straight head. One cue is the location of the iris within the lid aperture, so that a shift of the irises to the left signifies gaze to the left and vice versa.16,17,18 The other cue is the direction in which the iris appears to be “pointing.”17,19 These theories are not incompatible, and it has been argued that both cues are variably important under different conditions.19
Because angle kappa varies between individuals, it adds a variable error to both of the above cues. How it does this requires an explanation of how angle kappa is measured. In a typical setup, the subject faces a small light source and views a target along the line of sight (LS), the line which connects the target to the center of the pupil (Fig. 5). When the light source, the target, and the viewer are co-aligned, the viewer usually sees the corneal reflection of the light source (X) as nasally displaced from the center of the pupil. Then, the fixation target is moved nasal to the light source and viewer, with the eye and its LS following it, until the corneal reflection of the light source is centered in the pupil. This establishes the pupillary axis (PA), which is the line that goes through the center of the pupil and the center of curvature of the cornea, and the angle that the eye has turned is called kappa (K).20 Angle kappa is usually greatest horizontally and because PA is typically temporal to LS, a significant angle kappa tends to make the eye look exotropic.11
According to the pointing theory, the eye should appear to gaze in the direction of the eye's anterior-posterior axis of rotational symmetry. An external estimate of this would be the line that goes through center of rotation and the center of the iris, which would be approximately perpendicular to the center of the iris (the iris axis, IA). Because, the pupil is usually nasally decentered,9 this would give a discrepancy between the PA and the IA because the fixation target would have to be moved farther nasally to place the corneal reflection in the center of the iris rather than in the center of the pupil (angle delta10 rather than angle kappa). Therefore, even a 0-angle kappa would leave a significant temporal deviation of the IA from straight gaze. Even though this pupil displacement may seem small, it would result in a large difference between angle delta and kappa. Therefore, the IA should be directed closer to the direction of gaze than the PA (especially in eyes with dark irises). This has been suggested previously, but it has not been adequately tested.10,21
For all iris conditions, straight binocular gaze (0.0 cm) was judged to be a few centimeters to the observers' right [Fig. 3 (A to D)]. This was probably because the model's head and lids were not perfectly left-right symmetrical, which gave an uncertainty to the straightness of the head and a consequent turn of the eyes in the opposite direction.
In a previous study, straight monocular gaze was judged to be abducted from its true direction, and it was suggested that this was, at least in part, because the models had large uncorrected positive angle kappas.10 However, the particular angle kappas for those models could not explain all of that difference, and it was argued that the discrepancy may have been because the IA (the line perpendicular to the center of the iris) was a better indicator of the perceived direction of gaze than the PA was.
Because this study simulated a 0-angle kappa, it provides direct evidence that kappa alone cannot explain why straight gaze is usually perceived to be exotropic. If the pupil were centered in the iris, angles kappa and delta would be identical, and equal gaze directions would be predicted whether the pupil was poorly visible (dark iris) or easily visible (light iris). However, because most pupils are decentered nasally, angle delta would usually be larger than angle kappa. Therefore, to the extent that the pupils are decentered, gaze from eyes with pale irises (usually blue) should more closely follow the PA, whereas gaze from eyes with dark irises should follow the IA so that blue eyes should be perceived to be looking less far to the side than brown eyes. Fig. 3 shows this result. Oddly, when the blue irises had no pupil or a centered pupil (Fig. 3B, C), the divergence was greater than for the brown iris (Fig. 3A). This may be because the decentered pupil within the brown iris was still somewhat visible (Fig. 1B).
Previous studies have disagreed on whether binocular gaze to the side gives the perception that the gazer is looking farther or less far to the side than he or she actually is.5,10,17,19,22,23 In this study, binocular side gaze was overestimated. This disagreement within the literature does not seem to be explained by differences in the distance of the model or the target from the observer, or by differences in the extent of side gaze that has been tested. It is possibly because of differences in the eyelid shapes of the models, which have been shown to influence the perceived amount of eye turn.3
Studies have more consistently reported that the judged direction of binocular gaze follows the judged direction of monocular gaze from the abducting eye.5,10,19 This study supports this. The judged direction of straight binocular gaze seemed to be based on the average of the curves for straight monocular gaze (0.0 cm), but, for side gaze, binocular gaze tended to follow the judged monocular gaze of the abducting eye.
For all the simulated irises, the binocular curves had a steeper slope than the monocular curves which, except for the one in which highly visible pupils were decentered, caused the monocular curves to roughly coincide with the binocular curves at the true gaze locations of −3 cm and +3 cm (Fig. 3A to C). Because these crossing points agree with the half distance between the centers of the average pair of eyes, the judgment that binocular gaze is toward one of the observer's eyes agrees with that of monocular gaze from the eye on the same side. This may have some functional significance. Because gaze in face-to-face encounters is usually directed toward one or the other of an observer's eyes, the monocular/binocular crossing point would minimize any conflict between the perceived directions of monocular and binocular gaze. Because the eye to the same side is relatively abducted compared with the opposite adducting eye, binocular gaze may follow the abducted eye for other gaze angles as well to be consistent. The above applies to all brown eyes that have poorly visible pupils and to pale eyes that have little or no pupil decentration.
The monocular curves from the highly visible decentered pupils essentially overlapped each other and joined the binocular curve at 0.0 cm (Fig. 3D), so they did not show the distinct crossing points with the binocular curve mentioned above. The fact that the irises with easily visible decentered pupils had a greater agreement between monocular and binocular gaze over the entire observer's face than the other irises did may, in part, be the reason why such eyes can have such a piercing stare.
It might be assumed that the perceived direction for binocular gaze is the average of the two directions for monocular gaze, but this only occurs when gaze is straight. For side gaze, the function must be more complex because, even though the monocular curves differed for the different iris configurations (Fig. 4 OD, OS), all the curves for binocular gaze overlaid one another regardless of iris configuration (Fig. 4 OU). It is possible that gaze perception may involve a process similar to face recognition, which has been shown to involve more than the sum of the individual facial features.24
Finally, we might note other consequences of a nasally displaced pupil. The displacement of the pupil from the average optical axis of the eye has been shown to introduce monochromatic aberrations, which would degrade the retinal image,25 although it is unclear how a positive angle kappa might interact with a nasal pupil decentration to affect those aberrations. However, a nasal pupil decentration in conjunction with a positive angle kappa has been shown to reduce the effect of chromatic difference of magnification, which would improve the retinal image.21 The fact that nasally decentered pupils appear to result in a more accurate assessment of another person's monocular gaze may add an additional benefit of decentered pupils to this list.
This study was supported by a Northeastern State University Faculty Research Council grant.
Roger W. West
Oklahoma College of Optometry
Northeastern State University
1001 North Grand Avenue
Tahlequah, Oklahoma 74464