An attempt to integrate cognitive sciences with art making.

Monday, September 1, 2008

super portraits

a "superportrait" is an image that is paradoxically more like the person than the person themselves

Superportraits: Caricatures And Recognition (Essays in Cognitive Psychology) by Gillian Rhodes (Hardcover - Feb 1, 1997). It's now out of print and expensive to buy, but it is in many psychology libraries.

morphing software
To understand superportraits you have to be familiar with what is called "morphing" software. This software, about ten different programs are available, some without charge, over the internet, will make a series of images seamlessly transforming one image into another with all the steps in between available to see. One of the classical illustration of morphing is of the transformation of the face of a beautiful woman at one end to the face of a cheetah at the other. In between are images that look somewhat like the cheetah and somewhat like the woman. Below are steps 5 through 16. Step 1, not shown, would just be the picture of face of the woman.

step one: using morphing software to make an average face

There are, of course, many uses for morphing software. In this study the goal is to morph a great many photographic portraits into one. The result will be the average image of all the faces.
Individuals do not have perfectly average faces, but from a large collection of portrait photographs about the same size, you can use morphing software to combine several hundred faces all together. What you get is an image in which every feature you can think of, for example the nose, will be the average shape of everyone's feature, everyone's nose. This average face is not beautiful or striking. It doesn't call attention to itself, but it is a pleasant-looking face.

step two: morphing the average face with the face of the subject sitting for a portrait
If we now use a new specific individual as a portrait subject and the average face we developed above, we can do two types of relevant morphing transformations. We can make the subject's face more average. The result is called an "anti-caricature". Or we can exaggerate the subjects features by moving the image away from the average. This is called, of course, the "caricature". Most people are familiar with extreme caricatures, but in this case we do not overdo the caricature distortion. If the nose is longer than the average, we lengthen it some more but not enough to be recognized as a distorted photograph. We make the nose longer but keep it still within what might reasonably be seen in real life subject. Another example might be making someone slightly more bald than they actually are.
We will now have three photographs of our portrait subject to play with:

the anti-caricature -- a more average face
the caricature -- a less average face than the subject actually has with more exaggeration
the real picture of the subject -- the actual face without any changes

Here is a set of anti-caricatures and caricatures of Rowan Atkinson, the comedian. The image is a little small. The top center is an unaltered photo, on the left the column is increasingly normalized images, and on the right is increasing caricatures.

step three: the critical experiment

This experiment was done by S.E. Brenner for her masters thesis at the MIT Media Lab and published in Leonardo 18, 170-178 in 1985 under the title "Caricature generator: The dynamic exaggeration of faces by computer".

We now take these three pictures and ask the portrait subject and people who know him or her well which of our three photgraphs is most like the subject, which photo is the best likeness. We also test to see how long it takes for the person to be recognized in each of the photographs.

1. The anti-caricatures, the face made more normal than it actually is, is much more difficult to recognize, taking 4 times longer to recognize than the caricatures. I conclude from this that any effort to "pretty up" a portrait there will be a certain recognition cost. If you have an ugly face to make a portrait of, you will loose likeness by toning down the ugliness.

2. But more impressive is that the caricature is recognized twice as fast as the veridical image, the image that was not morphed either way. Even the subjects think the caricatures resembles them better than the real photographs.

We, therefore, have with our caricature what has come to be called a "superportrait", one that we can truly say is "more like the person than the person themselves". I conclude from this that the artist's job is not to make a likeness of anything (portrait, landscape or still life) but a super-likeness.

The immediate question I suppose an artist would ask is whether morphing software would help painters. Well, it probably would given how much we depend on photography, but the problem lies in obtaining the average face from which to create the caricature. We would have to ask with whom our portrait subject is going to be compared. There are a lot of practical problems. The group of face pictures we are going to morph together would have to be adjusted to the group from which the subject comes, that is for age, gender, class and race. It wouldn't work to morph an old man's face to a norm that is made from young faces. On the other hand, there is not an infinite number of groups from which subjects come, so it is possible that someone might do it.

Failing to develop this sort of software, I think the lesson is that, when working from a photograph or a model, there comes a time when you should stop using them and just develop the painting itself with what has been said above in mind.

how do we explain superportraits being better than exact portraits?

I think that exaggeration (which is what is meant by superportrait above) is one of three factors in making a super-likeness.

1. Little detail: too much detail leads to the frozen, expressionless look so often found in realistic portraits. Soft edges, particularly at the periphery, help. Air brush art is all soft edges and works quite well. Some sliding off the exact color up or down the color wheel, pushing green a little yellow at the edges for example, also reduce detail and increase the viewers ability to see the image as real.

2. simplification: recognition works when features are omitted (a portrait without eyes, for example, can be identified as such and such a person) but fails badly if the artist gets a feature wrong.

"From a slight underdetermined drawing, where the ideas of the composition and characters are ... only touched up, the imagination supplies more than the painter himself could possibly produce; and we accordingly often find that the finished work disappoints the expectations that were raised from the sketch." Joshua Reynolds (1723-92)

3. exaggerations: The brain does not identify a face by matching the image in the eye to a photo album of snapshots of faces stored in the brain by, for example, putting one image over the other and seeing if the features line up. What the brain does is use information processing. In an oversimplified example what the brain has is a "program" with questions like "is the nose long?". If it is someone with a long nose, it goes to the section of the program on "people I know with long noses" and asks other questions to narrow down the identification. Such a program uses much less memory than a photo album would. Where exaggerations, less detail, and simplification come in to help is, if the nose is made a little longer than it actually is as well as easier to recognize as a nose because it is simple and does not have extra detail, the program can get through the "Is-this-a-long-nosed-person?" step quicker. From the point of view of the brain, a better likeness is one it can identify quicker and easier. Interestingly Degas always painted his self-portrait with a longer nose than he actually had (we have photographs of him) and it is not clear whether he thought he looked more aristocratic or intuitively knew it was a superportrait.

There are some interesting historical facts. It is not that we have just discovered the useful of exaggerations.

The word "cartoon" is often used today in a way similar to caricature. However, most artists know the word as a preparatory drawing that will be used to transfer a composition to a wall for a fresco. It comes ultimately from the French and Italian, "carte", for paper and the italian "cartone" meaning pasteboard which is thicker than usual paper close to what we call "card stock" now. It's use in the modern sense comes from a humorous drawing from a 1843 Punch Magazine which depicted the House of Parliament fresco competition which was done by judging the cartoons submitted by the artists. But the drawing was itself labeled "cartoon" in Punch. The public apparently didn't know the meaning of the word, and from then on used it to refer to humorous drawings in magazines.

Deliberate exaggeration in an image of a person could only occur after there was a clear separation in people mind between the fate of the likeness of someone and the person themselves. Image magic is still one of the most widespread spells (burning in effigy, for example.) This freeing of the fate of the image from the person it depicted only occurred in the late Renaissance producing mannerism and caricature, a term coined by its first practitioner, Annibale Carracci.

"It [a caricature] is a likeness more true than mere imitation could be. And caricature, showing more of the essential, is truer than reality itself. This paradox is no means only an empty phrase...It reveals a new credo which is characteristic of the century in question". Ernst Kris.