Everyone is familiar with the witnesses to crimes being asked to describe the perpetrator's face. Whether it's a police artist or software, the usual approach is to attempt to assemble a remembered likeness feature by feature, what is called the "facial composite" approach. ("Did he have a mustache? Was it thick or thin?")
This is, indeed, one of the ways to go about making a portrait. Here in New York City it is most often seen in street artists many of whom were highly trained in China although one of my early teachers suggested this was the easiest way, and he was trained here. Usually one begins with one of the eyes. This distance between the eyes is usually the length of one eye, so it's easy to put the other eye in. One then measures from there to the other features in hopscotch fashion, and puts in the outline of the head last. You can get a quick likeness, but there is something stiff about it, something that is just not satisfying. In the case of the eyes, they sit on the surface rather than in the eye sockets even if after the fact attempts are made to put in lines suggesting it. One of the things you don't do in this method is squint because it obscures the features.
Well, it turns out that the facial composite approach to trying to get a witness to describe another person simple isn't much better than trying to draw a portrait this way. While the composite way of drawing a portrait is unsatisfying but recognizable, the facial composite software or artist's drawing of a remembered face is neither recognizable nor satisfying. When people were asked to make a composite of a well-known celebrity this way, only less than 3% of viewers could identify the person from the composite. The use of this faulty composite method is a serious problem. Of the first 180 people exonerated by DNA testing about 80% were falsely convicted based on witness misidentification.
Another approach is now being tried in police work that is called the "whole face" method. This one begins by showing people different types of complete faces -- square, round, heart-shaped, etc . The witness picks the "parent face" that most resembles the perpetrator from a display of about 5 other faces and then looks at different subspecies of this parent face until they can't get any further. This is beginning to look more promising for facial memory. Interestingly overall facial or head types are discussed all the time by hairdressers, and that's where you have to go to get more information about general head shapes. For example, the heart-shaped face with a small chin can have hair cut at the level of the chin to square up the facial outline a bit whereas the square-faced person with a strong, broad jaw would look terrible with this type of haircut.
Evidence is now accumulating that faces are processed, stored, and retrived at what is called the holistic level not at the level of individual features. This is probably the way we should draw and paint them. Artists who used to use a very out of focus slide to block in the head were probably on the right track. Returning to drawing or painting the eye, Sargent used to say that he painting the eye socket and then slipped the eye in like a poached egg at the end. In this method the more you squint in the beginning the better. I think it's a good idea to think of yourself as trying to first get the "parent face" painted. Instead of trying to draw the features I've been experimenting with (after I get some construction lines down) just drawing a smudge where the eye sockets should be, and if this is correct I might go back after a few other smudges like the shadow on the nose and try to define the eyelids and then the pupil.
(Association for Psychological Science (2007, Feb 11) Facial Composite Systems Falling Short.)
There is another recent study of visual short-term memory which I wrote about in my first post in this blog. The scientists found that, if people were given a little more time to look at them (4 seconds) they were able to remember more facial images than other things like cars and watches. On the other hand, if people were given a short amount of time (half a second), they did worse at facial memory. Once again the assumption is that faces are stored in a complex holistic fashion hence the need for extra time.
Most important and potentially extremely hopeful for us artists is that they think this supports the idea that short-term visual memory can be trained. That is we seemed to have learned to store faces better in short-term visual memory than other things, so why not other things or more of an individual face. James Gurney of Dinotopia fame, who has a great bog on art and art making, occasionally mentions that he has trained himself to be able to hold in short-term visual memory more and more of the subject I think by merely looking at the subject and looking away and trying to draw as much as he can. It's worth trying.
Vanderbilt University (2006, Dec 12) Why we "Never Forget a Face" Science Daily, May 21, 2008.