The first of these factors concerned the number of times the painter looked at his model to acquire the visual information.  During a small, 20x30 cm, preliminary pencil sketch that was completed in 12 minutes, he made 157 fixations on the model (6) at a rate of 13 fixations/minute, with an average fixation duration of 0.92 seconds.  For the final 130x100cm painting which was completed in about 100 hours over 12 days, we estimated that over 25,000 fixations had been made on the model, with a typical rate varying between 6 to 12 fixations/minute (after allowing for pauses for mixing paints, adjusting canvas, etc).  Average fixation duration in the painting case was slightly over 1 second although calculated only over short episodes of 5 to 6 minutes. Restricting our study at this stage to the drawing case alone, the artist was interrupting his work about 12 times a minute to look at the model for about 1 second each time.  Why did he not take longer looks at his subject matter and then spend longer periods reproducing it onto the canvas?

In trying to understand these timings, a fundamental rhythm of the painter's movements became apparent.  Each glance to model was the first element of a four-part cycle lasting a few seconds and which, in its complete form, consisted of:

1. Looking at a specific detail of the model;

2. Turning towards the picture;

3. Drawing or painting the detail;

4. Looking at the picture.

After this, the cycle was repeated for the next detail.

Steps 1 and 3 were the visual input and the manual output stages. Step 2 could be thought of as the main stage during which the brain processed the eye-hand transformation although, obviously, brain activity was taking place throughout the cycle. Step 4 was when the painter evaluated the work just done before starting a new cycle.

Depending on the type of drawing or painting and its state of progress, not all these steps were always present.  Thus, from step 1 and 2 the painter could return to 1 without drawing or painting anything, and after steps 3 and 4 the painter could return to 3 without glancing again at the model.  In very quick drawings, the 4th step would not take place, or rather, would take place during the actual drawing or painting stage.  In fact, there always was some degree of overlap between one step and the next, creating the fluidity of movement so beautiful to watch in a painter at work.  Finally, for longer portraits, step 3 could be preceded by a rehearsing action during which the hand practised a line many times before committing it to the paper, with the eye occasionally referring back to the model.  We shall return to this behaviour further on.

When the drawing or painting was finished, there were often long periods of repeated 1-4-1 steps during which the painter compared the model to the finished picture.  Interestingly, a similar behaviour was also sometimes observed before starting, when the painter seemed to "compare" the model to the blank paper or canvas.  Altogether, however, the full 1-2-3-4 cycle was the norm.

Thus, Double-Portrait enabled us to rough out the main elements of the picture production process.  However, in order to say more about the visual input-manual output transformation we needed to know more precisely what the eye was doing, i.e. where it was focusing, for how long, and what action the hand was performing in consequence.

For this we used an 'eyetracker' to record the movements of the artist's eye during short periods while he was drawing a portrait from life.  We also recorded his eye movements during selection of the model from a group of four candidates, and during a number of shorter portraits and sketches.  Some of the latter were done under strict laboratory conditions so that we could more carefully compare the artist's movements with those of untrained subjects.

For this project we combined these recordings with those from a hand movement sensor (Polhemus Fastrak motion analysis system), so that we could also follow the movement of the artist's pencil.  This device records the 3D position of a lightweight marker, attached to the artist's hand or pencil (7), with the data also digitally stored on the computer for later analysis. While these eye and hand movements were being recorded, the drawing's progress was filmed continuously with a close-up video camera.

A prime consideration throughout the project was to obtain data about an artist at work, rather than an artist performing scientific tests.  One of us (JT) had been filming Humphrey over a number of years in a non-interventionist style, and had devised with the artist a routine that allowed him to proceed as normally as possible.  As Humphrey was also well practised in drawing under difficult conditions, e.g. from moving cars or aeroplanes, the eyetracker did not constitute an undue hindrance.

Our observations were made as follows:

1. Selecting a model
The artist, wearing the eyetracker, sat facing an empty chair, and was presented one by one with four persons whom he had not seen before.  The duration and location of his initial fixations during the first few seconds were recorded in order to establish what he was looking for in a prospective model.

2. Brief sketches
The artist then wished to make brief sketches of his candidate models to help him in his selection.  He drew in ink in a small sketchpad held in his left hand, each sketch lasting 1-2 minutes.  Because of the difference in viewing distance between eye and model versus eye and sketchpad, these eyetracker results were used only to study fixation timings on the model.

3. The main portrait
Having chosen his model, the eyetracker system was set up on the following day to allow recording of the artist's right eye, as he sat with a near-vertical drawing pad positioned on an easel about 45cm in front and to the left of his head.  The model sat just to the right of the drawing pad, at the same distance, so minimising parallax errors in the video recordings made from the scene camera.  The eyetracker was calibrated and used for about 10-15 minutes each hour, and the artist then worked normally without the eyetracker for the next 30-40 minutes, while the model and artist rested for the remaining 10-15 minutes per hour.  Thus 5 recordings were made, spanning the entire process of the portrait drawing.  Humphrey's drawings were all life-size.

For the third and fifth session, we placed a motion tracking monitor on the back of the artist's hand, such that the position of the pencil could be calculated with an accuracy (8) of about 10mm, but with a resolution of about 1 mm.  This provided a record of the spatial location of major hand movements and also a record of the timing of all hand movement.

4. Secondary portraits
The following day, 3 short portraits were performed, two of a different subject and the third a self-portrait made using a mirror.  These drawings were aimed to take about 10 minutes each, so that the eyetracker could be worn throughout the entire process.  In fact, the second drawing took just under 12 minutes, while the self-portrait was completed in 22 minutes.

For each, a motion tracker was attached to the shaft of the artist's pencil, allowing him to use the pencil comfortably, but increasing the spatial accuracy of the recording of the pencil tip to about 2mm, again with a resolution of 1mm.

5. Comparison with untrained subjects
Some weeks after the detailed study, a series of very brief sketches were made by the same artist, and by three untrained subjects.  Each subject sketched from a black and white photocopy of a face (9), and was restricted to 1 minute per sketch. Again, both eye gaze and pencil positions were recorded.

1. Selecting a model
As each candidate in turn entered the scene camera's field of view, and sat down, Humphrey had already fixated on the person's left eye. He then proceeded with a number of rapid fixations, each lasting an average of 0.4 seconds, until his eye came to rest temporarily with a longer fixation of about 1 second (Figure 3).

With the last candidate, Nick (subsequently selected as the model for the main portrait), these first fixations were essentially confined to the person's eyes (10).  In this case, as in most of Humphrey's portrait work, empathy with the subject was the artist's overriding consideration.  It suggests that his eye movements during the first 2 to 4 seconds were governed by a first time encounter with a new face rather than by any professional consideration.  This high rate of fixations, about 140 per minute, was also seen when Humphrey viewed portraits produced by others, and when viewing his own completed works.  It is typical of the eye movement rate reported by others (11).

2. Brief sketches
Fixation rate and duration over the four sketches were remarkably consistent (Table 1). Overall average fixation duration on the model was 1 second, and the rate just under 22 fixations/minute.

3. The main portrait.
The data collected by the eyetracker can also be usefully viewed as a time series which emphasises the temporal pattern of eye movement (Figure 4). In the figures that follow, we have used the spatial location of the eye's gaze to colour code the time series, with the convention of lines shown in blue denoting the eye looking at the model, and in black, at the paper. Blinks, which hide the eye's pupil from the eyetracker camera, appear as vertical lines.

Humphrey's eye movements adopted a regular rhythm from the beginning of Nick's portrait (Figure 5).  For the first 35 seconds (Figure 5a), he scanned the blank paper occasionally referring back to the model, suggesting that a visualisation process was operating.  He then started with Nick's right eye (12), immediately falling into a pattern of regular fixations on the models face, each lasting about 0.6-1.0 seconds, at a rate of about 12 per minute.  During the five hours it took to make Nick's portrait, a total of about one hour was spent looking at the model, capturing each detail in epochs lasting about one second, the eye going directly to its intended target to "lock" onto it in a single stable fixation.  One of the first continuous lines Humphrey drew (at 55 seconds) was a 4.5 cm contour of the model's eyebrow (marked 5,6,7 in Figure 5b). While drawing this line, he stopped the pencil twice to glance back at the model, suggesting that his eye was capturing about 1.5 cm of detail per fixation.  The drawing of the right eye proceeded for about 1 hour with the accumulation of such small marks.

Some subtle variations of this basic pattern were observed over the five-hour period.  In Figure 6, we show 3 two minute segments of the process, the upper panel being the start of the session, the central panel occurring 1 hour later as he drew the hair, and the bottom panel taken during the 3rd hour as he drew the lips.  Notice the long fixations on the model at the start (the first minute); the very rapid sequences as he was drawing the hair and the regular and spaced sequence as he was drawing the lips.  These differences suggest that there may be a consistent relationship between the complexity of the visual object being viewed and the viewing pattern.  In particular, when considering the spatial location of the sweep of the hair, rapid comparison of the model and the paper was made.  When adding detail to the already partly drawn lips, a regular cyclic process of visual capture and then drawing took place.  The eye frequently returned to the same location on the model (13), at a rate that would indicate visual memory being refreshed on average every 5 seconds.

Towards the end of the session, an increasing number of fixations on the model were being replaced by fixations on the partly completed drawing.

Another observation highlighting the complexity of the drawing process is shown in Figure 7, panel 2.  When Humphrey started on the lips, the model's left eye had already been completed, but something was obviously not to the artist's satisfaction.  He therefore momentarily interrupted the drawing of the lips to return to the eye.  There he traced the drawn eye in a smooth cyclic pattern, at a rate of three cycles per second, with the pencil tip held just over the paper's surface.  These fine movements of the pencil tip were followed precisely by Humphrey's eyes, in a smooth movement (14).  Remarkably, however, no mark was ever made on the paper, and he eventually returned to continue the drawing of the lips.

The drawing of the lips was also frequently accompanied by such practice strokes (Figure 8).  The pencil would move several times just above the paper's surface, gradually defining the line before it was eventually drawn.  Occasionally, the eyes would look away with a rapid fixation to the corresponding detail on the model, or even to another part of the drawing, before returning to follow the pencil tip.  We suggest that the purpose of this behaviour is to aid in the control of the pencil.  Practice movements are seen in many tasks and sports requiring skilled movement, and probably allow the brain to refresh a short term 'motor memory' of how the body moves (15).  In other instances, the practice also resulted in very faint pencil marks being made on the paper, suggesting that they also aided in deciding on the exact form the final pencil line should take (16).  The differences between these two movement types, practice and drawing, were so slight (probably less than 1mm from the paper surface) that our tracking technique could not separate them and close-up video was required for their study.

Figure 10 shows the completed drawing, and the corresponding spatial record of the pencil tip movement. There must of course be a close correspondence between these two, but the differences reflect both the large scale motion of the pencil to and from the paper, indicated by the red segments of the line, while the differences between the black lines and the final drawing reflect the practice movements of the hand that were not committed to paper.

5. Comparison with untrained subjects
Finally, we present the preliminary comparisons that we have made between the behaviour of the artist, Humphrey Ocean, and three novices.  Here we imposed a time limit of 1 minute per sketch, and also requested the subjects to copy photographed faces, although neither of these is typical of Humphrey's style.  Figure 11 presents histograms of the distribution of eye fixation durations on the 'model' (the photograph).  The artist's fixation duration 0.6-1.0 second was typical of durations recorded previously.  The novices' durations were about half as long.  Furthermore, the artist's fixations were always single, whereas the novices' were generally multiple.  So the artist locked his gaze onto a single position, apparently taking in a single detail, while the novices fixated 2 or more positions, sometimes spatially quite separate.  The briefer and less consistent fixation durations of the novices were more typical of everyday eye movements, and more typical of Humphrey's fixation pattern when not drawing or painting.

Table 2: Humphrey Ocean's average fixation timings while drawing a portrait.

In contrast, while selecting his model, looking around the room or viewing paintings, Humphrey's eye movements followed a pattern of rapid fixations, each lasting about 0.3-0.4 seconds and repeated at a rate of about 140 fixations/minute. We refer to this pattern as his "normal" eye movements. The untrained subjects we tested, when asked to draw, did not show this change in eye movement behaviour from their normal pattern.

In between fixations on his model, Humphrey would look at his drawing with shorter, more rapid fixations, or, alternatively, with smooth movements which followed the pencil tip across the paper. Such co-ordination between eye and hand was most dramatically illustrated by the practice sequences when the pencil stroke was preceded by a rehearsing action just off the paper's surface.

When changing his gaze from picture to model, or model to picture, Humphrey's fixations unhesitatingly found, and "locked" onto, the minute detail he was targeting. This behaviour was in sharp contrast to that of the non-artists subjects we tested.

How then does a skilled painter transform a vision of the external world into a picture on the canvas?  The following five remarks help situate our observations of Humphrey Ocean in the more general context of his way of working.

1. The capture of visual information detail by detail, rather than in a more holistic manner, is reflected in the way the drawing or painting is built up.  This proceeds systematically, by small geometric areas, gradually building up to the picture's main elements: right eye, left eye, nose, lips, etc. Each detail and each element is of intrinsic importance.  In the artist's own words: "The shape you are putting down is always abstract. (…) That next rhomboid is the side of the nose - or it is an abstract shape. Each bit of the picture has to be able to exist in its own right. Even if there was none of the rest of the painting, if you put that bit down, it would work."

2. The artist's actions are essentially driven by the picture's progress - they are goal oriented rather than, as in his first encounter with his candidate models, stimulus controlled.  Thus his next glance at the model is for the purposes of advancing whatever he is drawing at the time. Such behaviour is quite different from the normal way of perceiving a face or a painting. Perception and cognition studies made in other areas, such as face recognition or the viewing of art, will not necessarily be relevant to the production of pictures.

3. The artist's eye and eye-hand skills are definable in terms of physiological parameters: fixation stability, fixation duration, targeting efficiency, etc.  As they are not found in untrained subjects, one may assume, at this stage at least, that they are acquired through training and practice, i.e. they are not innate.  Furthermore, with Humphrey, they are put to the service of precision. "If it (the line) lands a millimetre to the right or a millimetre to the left, it changes the weight, in some way, of the shape that it is describing. So when that line lands, you just want it to land in the right position, whatever that is!"

4. Nevertheless, the eye and eye-hand skills alone cannot define the picture production process.  Other artists working from life in Humphrey's style have similar skills and goals, yet, if asked to draw the same model, would produce entirely different portraits.  The reason for this is not how they draw, but what they draw.  There are very few lines in a portrait that represent actual lines on the face: a few wrinkles and perhaps a few contours.  The rest are subtle demarcations between areas of different light intensities, texture differences and colour changes.  Take for example the many lines drawn at the inner corner of Nick's right eye (Figure 12).  Not only are their precise locations on the model's face subjective to each viewer, but for the artist, they also depend on what he wants to do with them.  At the end of Double-Portrait Humphrey remarked about my hands that he had just painted: "I'm sure of what I am seeing, I'm not quite sure what I am going to do about it.  So I make a decision.  The final result is made up of a great many decisions."  The artist's skills allow him to draw with precision whatever line he chooses; the choice of line is determined by other factors.

5. The last observation centres on this choice: "At any given moment I will start from what I can see from where I am.  I try to achieve a likeness.  But what I want is a likeness to the reaction I have to something I can see".  Undoubtedly Humphrey's reaction is dominated by the visual input to his retina.  When he is doing a portrait, his vision concentrates nearly exclusively on the model and the canvas.  It is this vision that we have investigated with the eyetracker.  But with Humphrey, as with other painters, sound and smell are also very important, as are thoughts, preoccupations, memories of other portraits, even what he had for lunch, that tooth-ache, and, of-course, his artistic ambition at that precise moment in time.

Acknowledgements are due to the Wellcome Trust for support of Dr Miall's research and for a Sci-Art grant to the project.

Eyetracker - a head mounted device used to record eye movement, by providing a close-up video image of the pupil of one eye.

Fastrak motion analysis system. A device comprising a very weak electromagnetic transmitter mounted on a table top, and a small lightweight receiver mounted on the moving object, capable of recording the position and angle of a moving object within a sphere of about 90 cm radius.

Fixation - the eye moves most commonly in brief jumps, saccades, each lasting about 100ms. In between, the eye steadily views a single gaze position (or if the target is moving, a single object).

fMRI - functional magnetic resonance imaging. A technique to measure small changes in blood oxygen levels, that allows estimation of brain activity because of a close coupling between increased neural activity and increased blood flow to that area of the brain.

Gaze position - the position on the visual scene viewed during a fixation. In these experiments, we consider gaze position to include a single visual detail, although we cannot access by measuring eye movements alone how much peripheral visual input is also captured.

Resolution - the smallest observable movement of the eye; this is smaller than the spatial accuracy, as only the relative movement of the gaze position, and not its true position, is measured.

Saccade - the rapid motion of the eye from one position to another, usually lasting between 50 and 150 milliseconds, and typically occurring 2-3 times per second.

Spatial accuracy - the accuracy with which gaze position can be calculated, with respect to the actual gaze position.

6  Strictly speaking, we could only be sure at this stage that the artist was looking at his model, an action that we termed "glance".  However, subsequent eyetracker work with Humphrey showed that, virtually without exception, each of his glances was made up of a single fixation, and we can therefore anticipate here by using the term fixation.  Back to text

8  See glossary for explanation.  Back to text