Studies in Perception and Action VII S. Rogers & J. Effken (Eds.)! 2003 Lawrence Erlbaum Associates, Inc. The Mona Lisa Effect: Perception of Gaze Direction in Real and Pictured Faces Sheena Rogers 1, Melanie Lunsford 1, Lars Strother 2, & Michael Kubovy 2 1 James Madison University, USA 2 University of Virginia, USA The eyes in a portrait often seem to follow observers as they pass (the Mona Lisa effect). All 3-D objects in a picture, not only gaze, will rotate in virtual space as the observer moves past the picture (Rosinski & Farber, 1980). This phenomenon is predicted by the geometry of pictorial space (See Rogers, 1995, for a review) but it may also be due to limits in our ability to perceive the direction of another s gaze even in the real world, or to general inaccuracies in picture perception. Sedgwick s (1991) analysis shows that the virtual orientation of objects is affected both by the objective orientation of the gaze (towards the station point or away to one side) and by the degree to which the picture is slanted relative to the observer. According to the geometry, objective gaze direction should be increasingly mis-perceived (distorted) as the angle of gaze increases away from the station point (or center) (a differential rotation effect). We conducted two experiments to investigate the Mona Lisa effect as it relates to our ability to judge the direction of another s gaze. In the first experiment, we compared judged gaze direction for a real and pictured face, without slanting the picture plane. In the second experiment we measured the effects of slanting the picture plane on judgements of gaze direction. Method Experiment 1 Observers judged the direction of gaze of a live model and of life-size photographs of the same model. Only the eyes moved, the head
20 Rogers et al. was frontal. The observer was located at the station point for the picture so the optical geometry of the two conditions was matched. Figure 1.The apparatus and viewing arrangements for Experiment 1. The model (both live and photo) directed her gaze at 21 targets, placed horizontally on a panel positioned between the observer and the model/display, at 10 cm increments from the center (0 cm) to 100 cm, both to the left and right. Twenty-three observers participated in 2 blocks of the 21 randomized trials for both live and picture conditions. They indicated perceived gaze direction by shining a laser pointer along the panel to mark the location at which the model appeared to be gazing. Measurements were taken to the nearest.5 cm from the center of the panel to the point of light. Results Performance was similar for both the live and photo conditions. For targets located between 50 cm left and 50 cm right, no significant difference was found between live and photo conditions (see Figure 2). Judgments of gaze direction tended to be more accurate when the target was close to the center of the response panel, and less accurate at the extremes, especially in the photo condition. Figure 3 shows the 95% confidence intervals for the center target only (model looking straight ahead). Notice the greater variability in the photo condition. In figure 4 we show results for the targets between 10 cm and 40 cm to the left, which demonstrate that, in this range, gaze direction tends to be overestimated and that there is a small additive error for pictures.
Studies in Perception and Action VII 21 Perceived location of target (in cm) Actual location of target (in cm) Figure 2. Perceived gaze direction (indicated location of target along panel, in cm from the center) varied as a function of actual gaze direction (actual distance of target from the center) in both live and picture conditions. Perceived location of center target Figure 3. 95% confidence intervals for responses to the center target, showing greater variability for pictures. Figure 4. Perceived location of each target as a function of actual location (in cm). Observers tend to overestimate gaze direction, especially in the photo (top line).
22 Rogers et al. Experiment 2 We performed a second experiment to explore the effect of a slanted picture plane. In natural picture-viewing settings observers are rarely at the picture s station point. Slanted picture planes are the norm. We asked whether the Mona Lisa effect is limited to the situation when the model is looking directly at the observer (toward the perspective station point or camera), or if it is more general. For example, if the gaze of a picture is directed at your shoulder when you are at the station point, does her gaze follow your shoulder as you pass the picture? The geometry predicts that it will not. If the differential rotation effect is observed, gazes at the center target should appear to be towards the observer at all levels of screen rotation (the Mona Lisa effect). Gazes towards more extreme targets should appear to be increasingly rotated away from the target Method Observers judged gaze direction using a subset of the photographs from Experiment 1. We reduced the range of targets to 0 cm to 50 cm to the right and left in an effort to decrease the amount of off-board responses for extreme gaze locations. The computer monitor was rotated by 0, 15, 30, and 45 to the left and right. Four observers participated in multiple sessions of two blocks of 42 randomized trials (seven rotations by six targets). Responses were measured using the same apparatus and procedure as in Experiment 1. Figure 5. The apparatus and viewing arrangements for Experiment 2.
Studies in Perception and Action VII 23 Results Experiment 2 demonstrated the classic Mona Lisa effect. A model looking toward the station point appeared to be looking at the observer (a distortion) when the picture was slanted. The box plots in Figure 6 show responses for one observer for the center target at each level of picture rotation. Median responses are close to zero (the model appeared to be looking toward the observer), although variability increased as the display rotates further away. Perhaps because of an asymmetry in the model s face, the gaze appeared to drift away from the observer with display rotations to the left, although not with rotations to the right. Perceived distance of target from center 45 35 25 15 5-5 -15-25 -35-45 -45-30 -15 0 +15 +30 +45 Rotation of display in degrees (- to the left, + to the right) Figure 6. Box plots of responses to the center target, at each level of display rotation, for one observer. When the model was not looking toward the station point (targets at 10 50 cm from the center) the model s gaze appeared to rotate so that she seemed to be gazing at increasingly extreme locations on the panel as the picture was increasingly slanted away from the observer. There is a hint of differential rotation effects over this range of targets and display rotations, but it seems likely that such effects will become obvious only at extreme values of each variable.
24 Rogers et al. Discussion Experiment 1 showed that, for the perception of gaze direction, a photo can act as a reasonable substitute for a real scene when the observer is at the station point (center of projection for the picture). Even in this privileged viewing position, however, there was more variability in perception of the pictures and there was a consistent error in picture perception in the form of overestimates of the distance of the target and thus the gaze direction. When the observer is not near the station point for the picture (as is typically the case in picture viewing) systematic distortions in the perception of the objective pictured scene will occur. Rotation of three-dimensional objects, including gaze direction, in the virtual space of slanted pictures is predicted by the geometry of pictorial space. The results of Experiment 2 indicate that perception is governed by this geometry, and not by the geometry of the objective 3-D scene. The Mona Lisa effect is a unique feature of picture perception. It is an example of the predictable, systematic distortion in the perception of pictorial space that occurs under normal picture-viewing conditions. The implication for experimental psychologists and anyone else who uses pictures is that pictures should not be assumed to be adequate stand-ins for real scenes. More research is needed to map the distortions in pictorial space, and to identify situations when veridicality of perception can be expected. References Rogers, S. (1995). Perceiving pictorial space. In W. Epstein and S. Rogers (Eds.) Handbook of perception and cognition: Vol. 5. Perception of space and motion (2nd ed., pp. 119-163). San Diego: Academic Press. Rosinski, R. R., & Farber, J. (1980). Compensation for viewing point in the perception of pictured space. In M. A. Hagen (Ed.), The perception of pictures: Vol. 1. Alberti s window: The projective model of pictorial information (pp. 137-176). New York: Academic Press. Sedgwick, H. A. (1991). The effects of viewpoint on the virtual space of pictures. In S. R. Ellis (Ed.), Pictorial communication in virtual and real environments (pp. 460-479). New York: Taylor & France.