Friday, December 17, 2010

Proposal for a New Element of Design

Here is a long post, because I don't think it's fair to bring up this topic without treating at least all of the basics of the concept. But please don't feel obliged to read it all at one go - I just want you to be assured that I'm not leaving out anything you need to get the gist of the idea here.

This is another interesting concept resulting from my badly mis-remembering something I read one time. What I thought I remembered was this: that some scientist had done a quantitative analysis of Jackson Pollock's paintings and figured out a metric of information density on the paint surface.

The upshot of this was that it became possible to authenticate possible Pollocks by checking whether they stayed within Pollock's characteristic information density range.

Moreover, I remembered this information density metric as having low numbers, possibly between 0 and 1 or 1 and 3.

So, anyhoo, I was wrong. The scientist guy was University of Oregon physicist Richard Taylor, the information density metric was actually a fractal measure, and the numbering system was 0 to 3, with Pollock moving from 1.45 to 1.72 over the course of his mature career.

Oh, and apparently the authentication part doesn't really work. Here are a couple of articles, if you're interested, the first rather rah-rah, and the second distinctly less so:

Discover Magazine: Rah-rah!

Artsfuse: Blah-blah.

What was my interesting idea? Well, I'll tell you. I was over at my painter friend Andrew Smenos's studio the other evening, looking at his work, and I was trying to describe a difference between some passages in his paintings and other passages. I couldn't articulate the difference in terms of any of the standard elements of pictorial design.

Well, let's back up a second and go over those elements, just in case you've forgotten. I know I have trouble remembering them all:

Those are the ones I learned in school. They're like the earth-air-fire-water of art.

As I was saying, I couldn't articulate the difference between the passages in terms of any of these elements. But there was a difference. And that's very interesting, because these elements are supposed to provide a complete list of the available dimensions of definition of the visual field. That's the whole point of making up a list of elements.

My mind returned to my badly garbled memory of that Jackson Pollock study, and I said, "The difference is information density. There's low information density here, and there's high information density there."

What do I mean?

Well, look at this painting by Andrew, called Don't Be Such a Villain (oil on canvas, 48" x 36"):

That doll has low information density. Its surface is quite smooth and the gradient from light to dark is continuous.

The figure, in contrast, has high information density. There are abrupt transitions at a very small scale between light and dark, as well as an overall complex structure which, at a larger scale, also involves multiple abrupt light-dark transitions.

This concept of information density is similar to Dr. Taylor's fractal metric, but not exactly the same. It does not require repetition of the pattern at multiple scales, only a large amount of distinguishable information per unit of area:

Fractal scale - this is a graphic produced by Dr. Taylor

Information scale - I cooked this up myself.

The concept is similar to texture, but not the same as texture. One can imagine, for instance, depictions of dog fur and oak bark as having equal information densities, but distinct textures. Similarly, one can imagine conveying the texture of a leather chair by means of a low-information-density, smooth idiom, or a high-information-density, impressionist idiom. The texture is the same, but the information density varies.

So here's what I'm going to have to do. I'm going to have to propose that we add information density to our list of the fundamental elements of picture-making. You heard it here first.

Well, I'm sure there are lots of things that you could propose adding to the list of elements (although I can't think of any just now). Why is information density special? Why is anything special? Here's my list of qualifiers:

1. Because it's ubiquitous but aesthetically controllable.

2. Because it's useful.

3. Because the list of elements of design is noticeably, functionally incomplete without it.

4. Because artists interact with it in a structurally and cognitively similar manner to their interaction with the other elements of design.

I think we ought to put this to the test by looking at a couple of artists and the role that information density plays in their work.

First, since we're already getting acquainted with him, let's take a closer gander at Andrew Smenos's paintings.

We'll go back to the first one we looked at, Don't Be Such a Villain.

This seems as good a place as any to offer a rigorous definition of information density:

Information density is the number of visually distinguishable regions per unit of area.

[Aside to any information theorists in the room: I'm pretty sure I'm not using the term "information" in any Shannon-derived sense. What I mean by visual information is "visually discrete homogeneous region." For instance, the following items, in this poorly-thought-through and ill-defined system, would be considered single items of visual information:
  • a single white pixel
  • a circle filled with a uniform pale blue
  • an entire computer monitor showing nothing but white
  • a straight black line
  • a curved black line without inflection points
The following items would be considered two items of visual information:
  • a computer monitor showing nothing but white, except for a single black pixel
  • a white pixel beside a red pixel
  • a curved line with a single inflection point]
Back to the point. In Andrew's painting, the figure has high information density, the dolls have low information density, and the background has an information density of 1 - that is, the information content of mere existence.

If only we had some kind of information densitometer to test whether this intuitively clear classification is empirically valid! Wait a minute, we totally do. Well, a rough one anyway. You ever notice how different jpeg files of equal pixel count are unequal in file size? That's a function of information density. The jpeg compression system, like all lossy compression systems, doesn't store all the information in the source data set. If you think of all the information you need to define a pixel of image as having a data value of 1, then the information:pixel-number ratio of a jpeg is less than 1:1. How much less depends on how complicated an image the jpeg is compressing.

So let's try out our theory of information density in Andrew's painting using a simple trick: making new jpeg files of parts of it in Adobe Photoshop:

Each of the three regions is a square 85 pixels on a side. The jpeg file sizes are more useful as relative measures - relative to each other, for ranking - than as absolute measures. The picks I made here were my instinct for "typical" regions in the three parts of the painting, but I did put my thumb on the scale a bit: I chose a doll area with some object differentiation, and a figure area with very little form differentiation. The figure area *still* came out more information-dense. This effect is even more striking in person: the doll is uniformly smooth, and the figure is uniformly complex. This is a function of the difference in painting techniques used in painting the two objects.

Now, let's look at a few more Andrew paintings:

Mama Tiger, oil on canvas, 40" x 32"

There Used To Be So Much Anxiety, 36" x 24"

We see Andrew repeating the strong distinction between regions of low information density and high information density, and frequently including regions of information density 1 - the pure whites. And those whites, in person, are really, really white: he puts on so many smooth layers of paint and gesso that his canvas surfaces ultimately have a tight, hard, drum-like consistency. This is his intuitive method for reducing information density; it produces a uniform surface without even the texture of the canvas threads themselves to break up the white field.

So what does this analysis buy us, in relation to my criteria for considering the relevance of information density as a fundamental element of design? Here's the list again:

1. Because it's ubiquitous but aesthetically controllable.

Check. Andrew's painting has an information density distribution - and in fact, all paintings do. You might argue that on this basis, "visual information" alone should qualify as a fundamental element of design. And it's true that information is ubiquitous. But saying "a picture has visual information" is equivalent to saying "the picture exists." Information exists because the image exists. All visual objects have an information value of at least 1 (using my peculiar definition of information). Thus, the presence of information cannot be controlled - the artist cannot decide anything with regard to the mere existence of information, just as the artist cannot decide anything with regard to the existence in the work of line, shape, color, value, and texture. But the artist can decide on the distribution of information density over the space of the image; and this makes information density, unlike information alone, an aesthetic design element, while also ubiquitous (like the accepted list) because the mere existence of the picture brings information density into being.

2. Because it's useful.

Check. I think it's fair to argue that we have a better understanding of what we're seeing, when we look at Andrew's paintings, after our analysis than we had before it.

3. Because the list of elements of design is noticeably, functionally incomplete without it.

Check. Absent this analysis, there would be a major categorical absence in our formal understanding of Andrew's paintings. In case you didn't already think of this, let me draw it out a little farther. Compare the colors in Andrew's paintings, particularly this one:

You see what he's doing there? Try parsing this beneath the level of object-identification: forget for a second that one of the things visible here is a naked woman (it can be done with practice). Look at the picture in a mechanical sense - a purely formal sense. The eye is naturally drawn to certain extremes among the elements of design. The eye is drawn to more saturated colors over less saturated colors. It is also more drawn to high information density than it is to low information density. In this painting particularly, Andrew has located the regions of greater color saturation (the dolls) and the regions of higher information density (the figure) in different spaces in the painting. This produces visual tension because two different brain functions compete with each other when we look at the painting.

Perhaps you're thinking, "What does all this have to do with the humanity and meaning and so forth?" Easy there - in this case, that's not a fair question. This formal kind of analysis is what we do when we learn about the elements of design. The humanity part comes later. Back to the point: within the accepted five-element schema, no strong formal tension exists in Andrew's painting. But a strong formal tension definitely exists in our experience of it. So the five-element schema is brutally incomplete in its treatment of this work. At least as far as Andrew's paintings go, information density meets the second criterion for relevance.

4. Because artists interact with it in a structurally and cognitively similar manner to their interaction with the other elements of design.

Check. Previously in this blog, somewhere, we've talked about the distinction between form artists and color artists. We've probably also talked about the distinction between analytic colorists (like me) and organic colorists (like John Singer Freaking Sargent). These are examples of different types of interaction of different artists with the elements of design. In case you've forgotten, let me review the difference between an analytic colorist and an organic colorist, because it's a good example of what I'm talking about with point 4 here.

An analytic colorist decides on the colors of his painting largely in advance, according to an explicit design scheme. People who don't naturally think in terms of color tend to do this, ironically lending their paintings a heightened sense of color:

That's me, deploying an analytic color scheme.

And this is The Tragedy, by Severus Snape, painted in a haze of grief shortly after his betrayal of James and Lilly Potter. Snape was an analytic colorist.

Organic colorists, in contrast, focus much more intensely on depicting the colors in front of them, and are quite good at mixing and juxtaposing colors which are faithful to the real experience of color, closely examined.

This is a painting by Sargent catching both the quality and color of a light source, and of a zillion different textures and colors reflecting light from the source. Sargent is a master of this, and has spawned any number of contemporary imitators, some of whom are also quite good. Sargent is better though - he's as good as the impressionists, even if his line-shape idiom is profoundly different from theirs.

What this distinction illustrates is that all real elements of design are subject to aestheticization based on differences in the deep interaction of each artist with the element in question. Botticelli is a line man, and so is Schiele, in a way that Rembrandt simply will never be. The distinct, single line speaks to Botticelli and Schiele in a way that it does not speak to Rembrandt. Et cetera.

So when I throw in criterion 4 in my list of necessities for a fundamental element of design, I mean that each element must hold, implicit within it, the same broad capacity for individual variation in its deployment and expression. This variation must be structurally and cognitively similar from element to element - that is, the element must arise at such a profound level that the neuro-cognitive individuality of artists imprints their use of the element, producing a variety of kinds of art bearing certain structural commonalities, including an organic-analytic distinction. When I say that Picasso is an analytic colorist and Sargent is an organic colorist, I'm showing that color qualifies as an element of design because color lives deep in our brains.

And I am saying that information density lives deep in our brains too. I am also saying that Andrew is an analytic information density artist. He has developed an arbitrary but aesthetically sophisticated scheme of information density distinction in his paintings, returning to tripartite high-low-1 regions in his compositions.

Now let's look at an organic user of information density, probably the most talented organic information density painter I know. His name is Fedele Spadafora. This is his painting Food of the New Masters 4:

What has Fedele done here? Well, it's interesting. He has placed the regions of high information density at the points the eye naturally perceives them:

In the reflected highlights in the liquid in the foreground, he has created small zones of high information density.

In the heap of food in the background, he has created a large zone of high information density. Notice that he has not created high resolution - it is nearly impossible to tell exactly what the heap of food consists of. But that it has great variety of color and contrast over a small area is clear.

Apart from these regions, his image has low information density - the tablecloth, the rear cup, the plate in the background, all are relatively flat and featureless.

Notice him doing the same thing again in this painting, Food of the New Masters 5, which shows, if anything, even more sparsity of regions of high information density:

I leave it to you to closely read this painting. For my part, I'd like to pick apart Fedele's application of his technique to a different kind of scene - people sitting on a train, in his painting Path Train:

In this painting, we have variation in information density distribution acting to lead the eye through the painting. We start with the hand of the foreground figure and the newspaper he is reading (or whatever that thing is). This region includes mottled gradients, high contrast, distinct shapes, and a range of color. The white line along the right side of the arm leads us down, then up, to the face of the foreground figure, another region of high information density - notice how the skin is defined by small but distinct patches of color. Value also varies in distinct regions, going from the edge light on the left, a high value, to the midtowns of the forehead, to the secondary highlights of the cheek, to the darker midtones of the stubble, to the darker values of the shadowed eye, nostril, and chin. All of this occurs in a region which occupies a small part of the total surface area of the painting. Like the hand/newspaper area, information density here is extremely high.

The eye moves from the face to the vent in the background, which Fedele has chosen to give a comparatively high information density, clearly defining the highlights in the circular metal components.

But he has chosen not to provide information density to the second sitting figure. While we have shape and highlight, the internal detail of this face drops toward nothing. There is some definition of form, but the paint application is slightly smoother than for the foreground figure, and the range of colors and values is restricted.

Yet this figure is comparatively well-defined compared with the regions which have truly low information density in this image - the ceiling, the back wall, the overhead light source, the window behind the figures.

This kind of distribution of information density is often called "focusing the image by means of selective detail." This terminology alone tells us that the classical list of elements of design is incomplete. The statement does not depend on any particular element of design. It does not describe line, nor color, value, shape, or texture. It describes their distribution. The distribution is a question of information density.

But I claimed earlier that Fedele's work exemplifies an organic sense of information density. What did I mean by that? I meant that he replicates the natural functioning of the human visual system in taking in a scene. The eye-brain system performs approximately the following procedure when presented with a new view:

1. make a rapid low-information-density overall scan
2. select items of apparent interest
3. return to items of interest and scrutinize carefully, raising image resolution in selected regions of the visual field

Don't believe me? Perhaps, then, you will believe… Harvard Medical School neurobiologist Margaret Livingstone! Yes, I understand this is a bald-faced appeal to authority, but her book Vision and Art: The Biology of Seeing, which I have cited many times on this blog, is still awesome, and you should still buy a copy and read it. On page 78, she discusses visual resolution (a slightly different concept from information density in an image itself), and even touches on the concept of scenic areas "rich in information," but she does not pursue this concept; it is peripheral to her line of reasoning. On page 79, she provides this excellent graphic:

In the top image, we see a painting of a man in a forest (by I.I. Shishkin). In the bottom image, we see where the eye of a viewer traveled as the viewer sat looking at the painting (in a clever experiment by Russian psychologist A.L. Yarbus). Livingstone's analysis concentrates on the meaning of the points of focus of the viewer: behaviorally important objects like people and foreground trees. She also raises the concept of relatively greater contrast in the areas the viewer focused on, and greater "detail." Detail here is taken to mean detailed representation. But detail, as I was just saying, is in this sense a kind of subset of information density. It is equally possible to produce information density with many distinct brushmarks in a small area, even if the brushmarks do not represent anything in particular. This is a large part of what is meant generally by the painfully vague term "painterly." Back to the point: when Dr. Livingstone says that the eye is drawn to high detail, I contend that she might as well say the eye is drawn to information density. In Shishkin's painting, the flat black regions of unlit forest got little or no attention; the more informationally rich trees and figure got more attention.

The painting itself is highly naturalistic. So what this experiment replicated was the natural progression of the eye over an unfamiliar scene. The eye identifies points of interest, and then "paints in" their detail by repeatedly scanning and sweeping them. The brain constructs an impression of the scene with an uneven distribution of information density. Density tends to be low in areas of nearly-uniform visual stimulus - even though it is possible to find information there if it is actively sought - and density tends to be high in areas of rapidly changing high-difference-value visual stimulus.

This is the natural information distribution of ordinary eyesight, and it is this process which Fedele has intuitively identified and heightened in his work. What makes the work distinct from most other selective-focus paintings is the degree of naturalism in the focus, and stylization in the information density differential across the entire image.

That is, Fedele does not choose to focus only on the points of thematic interest in Western art: he focuses on the face, and on the vent. This almost magically replicates the idiosyncrasy of the wandering gaze.

And he doesn't just let his low-density regions go lower-density: he lets them go nearly flat, with the merest hints of texture and variation. He pushes the low-density visual idiom to a nearly graphical limit, but his work never looks unrealistic, because his understanding of the functioning of the eye-brain system is so profound.

Fedele's work reads as astonishingly realistic, and yet it is in no way photo-realistic. The locus of Fedele's realism is not in the comprehensiveness of what is seen, but in the evocation of how we see it. His information density aesthetic is organic.

Having looked at Fedele's work and Andrew's work, I hope you can see what I mean about the aesthetics of information density. Both artists are masters at controlling a design element which is not even recognized yet. Although the element is unnamed, both have intuitively recognized its value as a tool of artistic expression. However, they have diverged in their use of the tool, following differences in their artistic vision and cognitive processes. Andrew has schematized the distribution of information in his paintings, using information density in parallel with his use of objects, as a semiotic tool. Fedele, in contrast, has studied natural human modes of perception and replicated them, with slight stylization, to heighten the feeling of organic realism in the scenes he paints.

I think this covers most of what we need for the time being.

I have a few more ideas on this topic I'd like to share with you, but surely they can wait. In another post, I'd like to discuss a difference I see between absolute information density, and effective information density. I'd like to discuss the influence of distance between viewer and object on information density (including an apropos quotation from Ruskin). And I'd like to demonstrate that this concept fulfills another criterion of good science - it has high explanatory value. I'll demonstrate that by applying the concept to analysis of work from Rodin, Lachaise, Rothko, Rosenquist, beginning art students, Matisse, and me. Stay tuned!


  1. Have you seen the Einstein/Marilyn optical illusion? Here's a link:

    It actually works better with a smaller version of the picture - this one is so big you may need to go into another room to be able to see Marilyn - but this link has an explanation of the underlying phenomenon, so that's the link I chose.

    The author defines the difference as high spatial frequencies vs. low spatial frequencies. I think this is pretty close to your "information density" concept. What do you think?

  2. Dani, great post. Sorry it took me a while to get through it.

    Fred, interesting contribution as always.

    I know a thing but I can't remember how I know. So I can't vouch for its accuracy. But sometimes in dense fog there are huge pileups of cars traveling at inexplicably high speeds, given the conditions. Turns out, the fog-induced absence of those sharp lines, or 'high spatial frequencies' from Fred's link, trick the brain into perceiving the vehicle as traveling much more slowly than it actually is traveling.

    Might not be as relevant as I first thought, but I'm always kind of traveling through my own fog, so...

  3. Fred, Ed - I am so grateful for your comments. I couldn't ask for better guys to share my thoughts with. Fred, I had come across that image, but never thought much about it before. Now I *am* thinking about it, but I don't think I have anything insightful to say about it yet. It's very interesting, and it may interact with the concept, although from a neurological perspective. This is sort of like how the boundary between color and value really breaks down from a neurological perspective - some of my early posts dealt with this issue, that we have unequal brightness perceptions of equally bright colors because of our neural response curves, so that "lack of light" and "dark color" are functionally equivalent. But in terms of the elements of design, we can synthetically keep the two concepts apart. I think this image interacts with information density in the same way, perhaps - at a level involving the mechanics of sight which the analytic statement of the elements is not supposed to be refined enough to comment on. At least my analytic statement of the elements...

    Ed, please don't apologize for reading this damn thing slowly! So far it's going over like a lead balloon, so your making it through at all is way awesome, from my perspective. The dense fog thing sounds very plausible. I was caught in a nighttime whiteout in a mountain pass in Montana one time, and I completely lost my sense of depth and, after a while, of up and down. I thought, "Wow, this is it, I'm gonna die." Fortunately, it was bumper-to-bumper traffic. I just steered in the direction of the taillights of the truck in front of me - the only part of it I could see - and inched forward at a few miles an hour. After some interminable period of anxiety, we all made it down off that horrible pass. Last time I ignore advice from locals at the gas station, such as "Stop here. You don't want to go up there tonight."

  4. I must say you have an original idea about the fundamental nature of art and you defend it well. Nice work. Try to get it out there even more if you can.

    Two questions coming from a non-artist here. At what point does low ID turn into high ID as the onlooker moves towards the painting? I'm thinking of the impressionists here.

    Also, can this apply to photography just as easily as art?

  5. Chris - Thanks! Will do (about getting it out there).

    With regard to your questions -

    1. That's a topic I want to write more about, and it's actually where the Ruskin quotation enters the picture. But this post was such a lot of work that I'm taking a little breather before I write even more about what appears to be a futile and relatively unpopular idea. You get worn down - I still haven't finished "Edges and Edge Detection," and I really only have one more post to write about *that*.

    2. Yes. And sculpture.

  6. After reading this article in the Sunday NYT , I went in search of more to read on this topic. Eventually unearthed this blog post. Very interesting.

    1. DS - thanks very much for stopping by here, and for sharing the link to the NYT article. I'm glad you enjoyed this piece. Be well, and drop in sometimes.