The importance of the center of the canvas has long been appreciated in art, as has the way the eyes as revealing the personality of the subjects of portraits. Is there a consistent placement of the eyes relative to the canvas frame, based on the horizontal position of the eyes in portraits? Data from portraits over the past 2000 years quantify that one eye is centered with a standard deviation of less than + 5%. Classical texts on composition do not seem to mention the idea that the eyes as such should be positioned relative to the frame of the picture; the typical emphasis is on the placement of centers of mass in the frame or relative to the vanishing point in cases of central perspective. If such a compositional principle is not discussed in art analysis, it seems that its manifestation throughout the centuries and varieties of artistic styles (including the extreme styles of the 20th century) must be guided by unconscious perceptual processes.
The importance of the center of the canvas has long been appreciated in art (1), as has the importance of the eyes as revealing the personality of the subjects of portraits. The question addressed in the study is whether there is a consistent placement of the eyes relative to the canvas frame, with specific reference to their horizontal position. One hypothesis is that the center of the face has a consistent positioning (in portraits where both eyes are visible). This center of symmetry of the face is often discussed in art analysis (1-5) and may be expected to be used as an explicit compositional primitive by artists trained according to such analysis. The second hypothesis is that one eye of the portrait dominates the compositional positioning in relation to the frame.
Fig. 1. Portraits of rhte Murray family over four centuries (from Bruce and Young, 1998) , illustrating the degree to which one eye tends to be placed on the center line, despite a variety of styles and head positionings over four centuries. The white line runs down the horizontal center and is broken where it reaches the level of the most-centered eye.
Fig. 1 illustrates the degree to which an eye tends to be set near the center vertical despite the vraiety of styles over the centuries. A survey of classical texts on composition has reveal only a couple of mentions of the idea that the eyes as such should be positioned relative to the frame of the picture; the typical emphasis is on the placement of centers of mass in the frame or relative to the vanishing point in cases of central perspective (1-5, 12-18). If this compositional principle is not discussed in art analysis, it seems that its manifestation throughout the centuries and varieties of artistic styles (including the extreme styles of the 20th century) must be essentially unconscious.
How general is the phenomenon of eye centering in portraits? In one sense, Western European art may be considered to belong to an integrated school of continuous interrelation between the artists, from master to pupil, over the centuries. Once art academies and museums were set up, artists could go and imbibe the styles of their forebears, providing an unconscious transference of effective design principles even if they were not discussed specifically or committed to print. In this context, it is exciting that an epoch of portraiture has recently been uncovered that is quite remote from the Western Renaissance tradition. These are the Fayyum funerary portraits of the expatriate Greeks in Egypt, in the first and second centuries AD. Arising from a vibrant intellectual community in Egypt that included many converts from the Roman repression of the Jews, these portraits have an astonishing freshness of technique and vivid sense of the subject depicted. They have been largely ignored until recentlybecause they were from an outpost of the Roman Empire and seemed provincial with respect to the centers of Greek culture, but did not qualify as Egyptian art from their country of origin because they were painted by Greeks.
Fig. 2. Four portraits from the Fayyum period of ancient Egypt, illustrating many of the same artistic properties as the classical portraits of later millennia, including the proximity of one eye to the center vertical.
The Fayyum portraits were painted as a substitute for the carved death masks in the mummification practice that was adopted from the Egyptian culture by these indigenous Greeks. The portraits were painted rapidly in tempura (egg-white paint), but seem to have been done long before the death of the subject, since most of them show the subject in the prime of life. The artists were evidently able to employ a full range of the tools of realism so prized during the Renaissance, including foreshortening, shape-from-shading (known to artists as chiarascuro), reflective highlights, cast shadows, subtle color gradations, and so on. The portraits were wrapped in place with the elaborate burial shrouds, forming a defined frame against which the eye position can be judged. As can be seen in the four examples reproduced in Fig. 2, in many cases one eye is positioned quite close to the center of the aperture of the shroud, contrasting strikingly with the carved death masks of the Egyptians, which are always completely symmetric. Although not all Fayyum portraits are as decentered as these examples, it is interesting that the head is often shifted away from being centered in the frame, as though it were more important to bring one eye close to the center. Clearly, this remote community of portrait artists conform to many of the same principles as much of later Western art, which seems to have developed in complete ignorance of their forebears.
Quantitative Analysis of Portraits
To quantify the relation between eye position and the canvas frame, the horizontal positions of the eyes were measured in classical portraits by all 165 artists represented in a variety of published summary sources drawing from the past six centuries (4-11). The portraits were selected as the first by a given artist in each source (4-11) meeting the following criteria: that the portraits were drawn by hand (oil paintings, watercolors, drawings or engravings) to ensure that the artist had maximum control over the composition; that there was only one person in the portrait; that both eyes were visible; and that depiction of the body did not go below the waist (to ensure that the head rather than the figure was the principal element of the composition). For comparison, the positions of the centers of the mouths were also measured. The single-eye placement hypothesis was evaluated formally by defining the most-centered eye of a portrait as the one closest to the vertical center line. If eyes were positioned according to the center of symmetry of the two eyes in relation to the vertical axis, both eyes will be about the same distance from the axis and the choice of eye will make little difference to the result. Conversely, if the head is positioned randomly relative to the center vertical, choice of the closer eye as the one for analysis will narrow the distribution somewhat, but by no more than a factor of , the standard deviation of the minimum of two samples from a Gaussian distribution.
Fig. 3 a. Histogram of lateral location of the most-centered eye (filled triangles) in 165 portraits over the past 600 years. Eye position was defined as position of the center of the eye opening, regardless of pupil position. b. Distribution of horizontal mouth positions (filled circles) is much broader than for the eyes in a. c. Distribution of horizontal eye positions in the 23 profile portraits from the same sources (defined as portraits in which the head is turned so much that only one eye is visible). The distribution (open circles) is even broader than the distribution of features in more frontal portraits.
The histogram of Fig. 3a vividly illustrates how one eye is placed in a narrow distribution peaking at the lateral center in portraits over the past 6 centuries (s.d. = + 4.2%). Other features of the face do not seem to be accurately centered. The horizontal position of the mouth, for example, is spread across the frame (Fig 3b), with a distribution about two-and-a-half times wider than that of the best-centered eye (s.d. = + 10.5%, significantly wider than 1.414 x 4.2%, p < 0.01). On the other hand, one type of portrait that is identified as not adhering strongly to the same principles is the side view of the head. In the smaller sample of side views from the same sources (4-11), the eye positions are scattered widely throughout the frame (s.d. = 18%, significantly wider than 1.414 x 4.2%, p < 0.01, even for this small sample). Evidently, passing to the pose of the side view changes the rule of composition so that eye-centering is no longer paramount. Perhaps, now that the subject's attention is perceived as being directed away from the viewer, the relevant principle becomes the centering of the head in the frame because the head dominates the composition.
Fig. 4. Analysis of size-scaled distributions of lateral eye positions in single-figure studies, which include a full range of head sizes. a. Expected lateral positions of the eyes on face-centered hypothesis fall on a positive slope (black line) on the assumption that face is centered (The left eye is flipped around the center line from its expected position on the dashed line). b. Eye-centering hypothesis predicts an additional centered bar of the most-centered eyes on the same assumption. c. Empirical positions of eyes in single-figure studies from 282 different artists over the past 600 years. Rightward eyes (sitters left eye) are shown as open circles, leftward eyes as filled circles, whose position is flipped around the center line. Note the absence of locations in the lower half of the panel away from the center axis, implying a distribution biased toward centering of one eye. If eyes were symmetric in the frame, flipping the leftward half of this distribution would not remove the symmetry. The pattern of distribution of all eyes is thus closest to eye-centered pattern of the scaled centering of one eye laterally, depicted in Fig. 4b, rather than the single oblique line of the face-centering principle of Fig. 4a.
It is interesting to compare the accuracy of eye placement with that in a psychophysical study of error in the placement of elements within a frame (19). Reproduction of the position of a single dot was accurate to about + 2% of frame width, while accuracy fell to about + 5% for the placement of a dot in a background of 4 or more dots. It therefore seems that the unconscious (or unexpressed) placement of the eye in artistic portraits (Fig 3a) is nearly as accurate as the attentive performance of those who were focusing on positioning on a background as their sole perceptual task.
Scaling of Eye Position With Scale of Portrait
One factor in the greater extension of the eye distribution vertically than horizontally may be the fact that the portraits had a range of scales, from those in which the head filled most of the canvas from edge to edge, to those that include the body down to the waist. It seems natural to move the head up in the frame as more of the body is depicted, to avoid having a space above the head equal to the extent of the body depicted. Such a principle would increase the vertical spread of the distribution without affecting its horizontal extent. (The tendency to move the head up in the frame as more of the body is depicted is not a necessary corollary, merely a plausible one. It could be, for example, that good composition called for the head to be centered vertically and the space above it to be filled with some relevant compositional feature for balance.)
To evaluate the role of such a scaling principle in eye placement, the range of scales was extended by adding from the same source books the set of single-figure studies that included portions of the body below the waist. This full range of figural studies now require evaluation of potential eye position principles, which will be structured in terms of the positions of the two eyes separately to remove the small narrowing of the distribution laterally by the choice of the most-centered eye in Fig. 3. The figure studies were selected according to the same principles as the portraits of Fig. 2, except that all figure lengths were included. Thus, only the first painting by a given artist was taken from each source, so that there is again only one painting by each of 282 artists in the sample (with its selection determined by serial order in each publication and hence independent of the present analyses). Reclining figures were, however, excluded, from the sample; only seated or standing poses were accepted. The rightward of the two eyes in the frame is shown by filled symbols, the other eye by open symbols. To provide a clearer analysis, the eye positions were all represented as if the mean eye position were to the right of center, by flipping all the portraits whose mean eye position was to the left around the center vertical.
To encompass the full range of head and figure sizes, the eye-position data are analyzed in a proportional plot. Lateral position in the frame is depicted on the ordinate as a function of the width between the eyes (on the abscissa). Thus, the larger the size of the head relative to the picture frame, the further to the right will the data be plotted. Predictions for the eye positions expected on two different hypotheses are provided in Fig. 4 a & b. If the face were typically centered in the frame, the mean positions of the rightward eyes would be expected to fall a positive slope (black line). The leftward eyes would fall on a negative slope (dashed line), but will be reflected onto the same positive slope by the convention of flipping the leftward-positioned faces. The width of the eye-configuration goes to maximum when the face fills the frame, assuming that the eyes are halfway down the head outline, so that the region of the plot beyond an eye width of 0.5 is expected to be empty. In short, the face-centering hypothesis predicts that all the data will fall close to the positive oblique in the flipped proportional plot.
The hypothesis that one eye is centered (Fig. 4b) predicts an additional horizontal bar of centered positions of the most-centered eye, with the other-eye positions falling on a positive slope (for both left and right eyes, after flipping). Thus, centering one eye would result in two swaths of data falling close to these two construction lines. If one eye is centered, the face must be situated to right or left, allowing a somewhat broader base to the distribution than when the face was centered.
The data of Fig 4c show that the eye positions are equally distributed between the 0.5 line of eye-centering and the positive slope, strongly supporting the hypothesis that one eye is being centered. The eye-position data thus go against the face-centering hypothesis, which would predict a lack of eyes located on the 0.5 line. The standard deviation of the most-centered eye distribution for the full range of figure studies was even less than for portraits (s.d. = + 3.7%). Conversely, the overall standard deviation of the other eye positions, which should have been greater by no more than if it were the face that were being centered, was larger by nearly a factor of 4 (s.d. = + 14.0%, significantly wider than 1.414 x 3.7%, p < 0.01).
The hypotheses of Fig. 4 are only two of a variety of hypotheses that could be drawn from current art analysis as to the scaled positioning of eyes in single-figure studies (in addition to the null hypothesis that there is such a variety of influences that the eyes will be positioned in a broad distribution that merely avoids their going too close to the edge of the canvas). These include the idea that the eyes, as foci of interest, might be placed at a Golden Section ratio laterally across the canvas and that the artists might specifically avoid the center as being too static a compositional structure. The results of the scatter plot of Fig. 4c conform to the one-eye centered hypothesis with remarkable accuracy, considering that this is a post-hoc analysis of art practices over the ages rather than a controlled psychophysical study.
In summary, the analyses presented reveal a dominant positioning principle for one eye to lie at the vertical axis of portraits over the past two millennia. The center of facial symmetry, which is the explicit organizing feature in many analyses of portrait technique, plays only a minor role in the positioning. The expression of this compositional principle with an unbiased accuracy of less than + 5% over such an extended time span is remarkable. The fact that this precision has been obtained from perceptual processes that appear to be completely implicit in the artists own analyses of their work suggests that hidden principles are operating in our aesthetic judgments. The ultimate significance of this particular placement of eye positions remains to be determined, but it seems that the window on the soul is most transparent in a restricted region in the picture frame. Placements of the eyes away from this region are usually used to depict indolence or some other reduction in our connection with the subject of the painting.
1. Arnheim R. The Power of the Center. University of California Press: Berkeley, Ca, 1988.
2. Arnheim R. Art and Visual Perception: A Psychology of the Creative Eye. University of California Press: Berkeley, 1954.
3. Jaxtheimer B.W. How to Paint and Draw. Thames & Hudson: London, 1982.
4. Edwards B. Drawing on the Right Side of the Brain. Tarcher/Putnam: New York, 1989,
5. Goldwater R. & Treves M. Artists on Art, from the XIV to the XX Century. Pantheon: New York, 1945.
6. Schneider, N. The Art of the Portrait: Masterpieces of European Portrait Painting 1420-1670. Taschen: Cologne, 1994
7. Petersen K. & Wilson J.J. Women Artists: Recognition and Reappraisal from the Early Middle Ages to the Twentieth Century. The Womens Press: London, 1978.
8. The Smithsonian Staff, The Smithsonian Experience. The Smithsonian Institution: Washington DC, 1977.
9. Coates T. Creating a Self-Portrait. Mitchell Beasley: London.
10. Wilson S. Holbein to Hockney: A History of British Art. Bodley Head: London, 1979.
11. Robinson J.M. Windsor Castle Official Guide. Michael Joseph/Penguin Books: London, 1995.
12. Alberti L.B. On Painting (1435). Trans. J.R. Spencer, Yale University Press, New Haven, 1966.
13. Ruskin J. The Laws of Fesole (1877-1879). Allworth Press: New York, 1996.
14. Henri R. The Art Spirit. J.B. Lippincott: Philadelphia, 1923.
15. Gombrich E.H. Art and Illusion: A Study in the Psychology of Pictorial Representation. Princeton University Press: Princeton NJ, 1969.
16. Kubovy M. The Psychology of Perspective and Renaissance Art. Cambridge University Press: Cambridge, Mass, 1986.
17. Dunn C. Conversations in Paint. Workman Publishing: New York, 1995.
19. Igel A. & Harvey L.O. Spatial distortions in visual perception. Gestalt Theory, 13, 210-231, 1991.