Color
From CS294-10 Visualization Fa08
Lecture on Oct 15, 2008
Readings
- Color and information, In Envisioning Information, E. Tufte
- A rule-based system for assisting color map selection, Bergman, Treinish, Rogowitz, (html)
- Color guidelines, Brewer, (html)
Demonstrations
Simon Tan - Oct 15, 2008 02:37:58 am
Some of the more basic rules of using color in visualizations (i.e. use muted colors instead of fully saturated ones, too little contrast is bad, too much contrast is bad) seem to hold in the realm of web design as well. There are countless resources that aid web developers in choosing aesthetically pleasing and highly usable color palettes where each color works well with those around it. For example:
For visualizations, Tufte seems to make the argument that color as an encoding technique is almost never enough on its own. He concludes the chapter on Color and Information by describing methods of adding redundant information to back up color encoding, as if dealing with color naturally means one must deal with the required redundant labelling or marking to accent it.
I found it interesting that the one visual variable that most people instinctively use in visualizations is the one inescapably in need of extra clarification.
NickDoty - Oct 15, 2008 02:07:40 pm
I particularly like these guidelines: it's comforting to have definitive answers to some of these issues. Also, the discussion of designing for color-blind users and which color combinations that are safe to use is interesting to see.
But I thought the most insightful were the comments in the Bergman paper about perception of color in contrast to "objective" measures like hue scale or RGB values. I've been confused in working on projects using a hue scale that the movement through the colors doesn't appear to be smooth: I thought there was something wrong with my algorithm or hardware when really this is an issue with our perception. It's similar with sound (and we discussed more examples like that during the perception lectures) that volume controls which work linearally sound like they're imbalanced (since they're not logarithmic).
Seth Horrigan - Oct 15, 2008 02:21:52 pm
The "Types of Color Schemes" represents a very succinct and potentially useful summary of a wealth of color wisdom. This information combined with tools like ColorBrewer (to indicate what color combinations would be most appropriate for a specific medium), offers a hugely useful introduction I would recommend to anyone who will be designing visualizations but have never had a course that covered color theory.
Brewer condenses many in-depth topics into a single sentence (e.g. "Different surroundings also change the appearance of a color."). I cannot say exactly how I feel about this; on one hand, it allows many topics to be covered quickly, on the other hand, excepting a critical reading of the paper and a reasonable amount of cogitation on each tidbit, many important details could be missed. I was particularly interested to see Brewer's take on aesthetics: "There has been a substantial amount of loosely structured research on color preferences. Regardless of context, it seems that most people like blue and don’t like yellow, but that is overly simplistic guidance for use in multi-color contexts." Still, all in all, I would highly recommend this to anyone who wanted an introduction to color theory. I think, however, that actual samples of the colors in question, rather than just text would make the page much more useful.
Kuang - Oct 15, 2008 03:30:39 pm
I never realized the concept of color goes so deep! Really enjoyed this lecture. And the link to ColorBrewer seems exceptionally useful.
One question is: any existing work on picking colors automatically with a learning system that takes efficacy to user tasks + prior distribution of values in a domain into account?
Nicholas Kong - Oct 15, 2008 05:44:05 pm
In response to the question brought up in class about the overwhelming amount of green in the standard color ramp, it may very well be perceptually based. Humans are biologically more sensitive to green; the exact reason escapes me at the moment, but it may have something to do either with more green photoreceptors in the eye, or because when looking at the eye's color sensitivity from the slides there is a large peak in the middle of the visible light spectrum, roughly corresponding to green.
I've done some work in image compression and camera imaging, and algorithms in these areas also make use of perceptual concepts. For example, most digital cameras contain only one photo-sensor array, even though color images typically consist of three planes (RGB, HSV, YCbCr, etc.). We are therefore undersampling the image. The most ubiquitous solution is the Bayer filter array, which contains 50% green sensors, and 25% each of blue and red, for perhaps the same biologically-based reason.
In image compression, JPEG makes use of a RGB to YCbCr color transform. YCbCr is an alternate color space in which Y roughly corresponds to luminance, Cb to blue chrominance, and Cr to red chrominance. In this space the Y plane contains most of the visual information. Therefore, the Cb and Cr planes can be more aggressively compressed without a commensurate decrease in the perceived quality of the reconstructed image.
Coming from this very different approach to colorspaces, I found it fascinating to learn about perceptually-based colorspaces and how these transforms could be applied to visualization.
Ketrina Yim - Oct 15, 2008 05:49:21 pm
The perception of color is such an important concept that it's covered in a number of fields: art, computer graphics, visualization, biology, and computer vision. I say this because I found it a little strange that both of the classes I had today discussed color perception.
I'm particularly intrigued by the possibility of devising an algorithm to generate color palettes for visualization. Having manually chosen palettes numerous times for both visualizations (namely Assignment 3) and artwork, I can say it can sometimes be a tricky task. Not only do you have to think of what colors maximize a viewer's ability to distinguish data points, but also in many cases the connotations of a color must be considered. For example, viewers could be confused by a temperature scale that uses red to represent "cold" and blue to represent "hot." Of course in some cases, such as stars, blue is hotter than red, but that's not what people initially think of. Then there are considerations that must be made to reduce eyestrain, prevent unintentional emphasis, and eliminate any chance of someone being unwilling to look at a chart simply because it's unsightly. An algorithmic approach to choosing colors would help those lacking design experience quickly select an effective palette for visualizations.
On a side note, after playing around with ColorBrewer, I've noticed that monochromatic palettes are the most colorblind-friendly, with the qualitative ones being the least. This I find hard to believe. There has to be a qualitative palette that is also compatible with the colorblind!
Scott Murray - Oct 16, 2008 11:41:21 am
I found this lecture immensely helpful in defining the perceptual attributes and relationships of sets of colors in concrete ways. There's a project I've been working on that currently uses randomly selected colors to group information. But after this lecture, now I'm motivated to write some code that, given certain constraints, would select appropriate, color-blind-proof, nameable colors.
Razvan Carbunescu - Oct 18, 2008 12:25:14 pm
One thing that I found interesting about this lecture was the fact that there was no discussion about the difference between additive colors and subtractive colors. I
additive colors (which are the colors used in TV's, computers, cell phones, etc. (base colors: red, green, blue))
vs
subtractive colors (which are used on canvas/paper/print to create colors (base colors: yellow,magenta,cyan))
While I realize that probably most if not all of us will only deal with additive colors in designing final project I just thought this was something worth mentioning.
Maxwell Pretzlav - Oct 19, 2008 11:08:05 pm
I would really like to try the color system described in A rule-based system for assisting color map selection, but it doesn't seem to be available anywhere. The system it uses is now open source as OpenDX, and I was able to install it via macports fine (and this software alone looks like it warrants some exploration). I got as far as this page at IBM, where the PRAVDAColor modules for Data Explorer are discussed, but mysteriously absent. Does anyone know if this is available, or is it just too old?
Chris - Oct 20, 2008 03:37:58 am
I was impressed with the results shown by PRAVDAcolor, especially compared with the default colormap results which I'm used to seeing (e.g, in Matlab), which made the fact that they gave such a brief description of the algorithm quite disappointing. I was surprised to see that the algorithm used was mostly spatially-based, as opposed to being based on the input data histogram, which is what is analyzed in most algorithms I have seen. I am curious as to how important that (the correspondence between value and color value) was to getting the better results that they demonstrated, versus just using a different target set of color values.
@Nicholas, I noticed this parallel too. To be even more emphatic about how aggressive the compression disparity in Y/Cr/Cb is, it is standard for the Cr/Cb planes of the image to be decimated by a factor of 2 compared with the Y plane even before any compression is done. This mirrors the disparity in sensors which you noticed on the data input side.
Michael So - Oct 22, 2008 12:19:33 am
The ColorBrewer seems like it would be pretty useful. I particularly like the legend indicating whether the color mapping is color-blind friendly, LCD projector friendly, etc. It was also interesting in the Brewer reading to discover that color is three-dimensional (hue, lightness, saturation). I didn't realize that before. And I am also now aware that there are three (four if you count binary as separate from qualitative) basic schemes when using color (sequential, qualitative, diverging...). I've seen these schemes applied before, but now I can give them names.
I also found it interesting that the perception of colors can change depending on the surrounding colors. Like the Bezold effect illustrated in the lecture slides. It's surprising that providing a dark border can make the those orange and green appear darker. It is an interesting effect and I am not sure of how to make use of it. Well, in the Tufte reading apparently it is applied in conventional road maps.
James Hamlin - Oct 22, 2008 11:48:42 am
So the demonstrations in lecture of human perceptual quirks were giddily fun, like seeing simple magic tricks as a kid. I was especially entertained by the demonstration of how perceived value of a color is dependent on its surroundings. This made me wonder how artists working in traditional media, like paints, develop a skill for using colors that produce the perceptually correct effect. I can imagine myself using colors with the perceived values rather than the actual values, and it seems like it would take some conscious effort to counter this.
Witton Chou - Oct 22, 2008 01:20:25 pm
I never realized there was such great research into the use of color. I would always try trial and error, but of course that can only account for so many factors. It is great to have some guidelines to work by to work efficiently and effectively in building visualizations. The use of color is important and if they can't be distinguished due to the nature of the medium on which it is presented on, the color encodings would be quite useless.
Matt Gedigian - Oct 22, 2008 11:11:55 am
Is Tufte right in saying that in all systems of color organization, "every color is located in three space." ... "A variety of color systems, but always three dimensions."? What about CMYK in which a fourth 'color' (blacK) is used? Some printers use six (or more) colors of ink to expand the gamut. Is this just because of differences in additive vs subtractive? I don't think so because Tufte includes CMY in his list of three dimensional color systems. It seems like even with additive, 3 dimensions give you a triangle of representable colors -- adding a 4th might increase the range (although you lose the unique representation for each color).
Jeff Bowman - Oct 22, 2008 01:50:27 pm
@Matt: Yes, Tufte is right, for purely biological reasons. As mentioned in the graphics undergraduate course (CS188), the eye has three types of color-sensing "cone" cells: Short cones, medium cones, and long cones. The long cones see red, the medium cones see green, and the short cones see blue. They respond to different frequencies as well, such the spectral orange color (from sunsets) that is not actually created by blending red and green.
However, by using red, green, and blue, a monitor can produce the same response from the short, medium, and long cones to approximate any given color. Therefore, any given human-perceived color has a single representation in three-space, and color systems like CMY have a unique mapping in 3-space. If our eyes had a fourth cone, like "orange", then it would respond most to orange light and we'd need to represent colors in four-space.
The use of black ink is just because it would be prohibitive to layer on enough black ink to produce a color that looks like black, for as often as we use the color black.
</optics-rant>
Dmason - Oct 22, 2008 01:57:09 pm
The choice of color palette for depicting quantum waves (and any complex field in R^2) is a particularly interesting subject. The de facto standard is to encode the phase of the wave in hue and the probability density 
as lightness, leaving the saturation at maximum for the entire image.
However, there are major problems with this depiction. For instance, where the wavefunction is small in magnitude, the phase becomes impossible to read. Some alternatives include just plotting the hue and drawing boundaries at some pre-defined amplitude.
And of course, you can always just plot the probability density and give it an arbitrary gradient map, as my advisor does for some of his artwork: [Eric Heller Gallery]
Yuta Morimoto
The perception of color is very interesting concept. Color aids to rapidly understand any visualization. Especially, I think the most effective usage in color scheme is appeared in charts or graphs. As we see in class, Tableau chooses the scheme of default color very carefully. Adobe Illustrator Flex also has the default color scheme for visual representation such as pie chart. I think both software choose default color, but surprisingly their choice resemble each other. I wonder that there is an accidental result or there is an optimal solution to color scheme.
Calvin Ardi - Oct 22, 2008 02:35:48 pm
The lecture on color has really helped me to understand a bit more about, well, color in general and some of the standards and guidelines for choosing and displaying colors. I've had some experience with the color gamut graphs with respects to monitors; a lot of times monitors (especially the cheaper ones) aren't able to display all the various colors that someone might need for a specific purpose (e.g., medical imaging, publications, graphic and video editing). It was interesting to find out that there is a standards body out there, and to learn about the differences between models for color. I've played around with RGB and HSV a bit, as well as CMYK, a color model used a lot in publications/printing.
It was also interesting about the role colors play, especially within context. Red/green has the meaning of stop/go, for example, but how are these colors treated in different cultures and communities? As Ketrina pointed out above, the red/blue pair has a hot/cold meaning, but are flipped in reverse in a diagram of stars. Is there some sort of manual or reading out there that establishes the semantics of certain color choices?
HeatherDolan - Oct 23, 2008 12:29:17 am
I enjoyed the color lecture a lot. Using color effectively is challenging.
Something important to consider with respect to RGB and CMYK color spaces is that the color gamuts are quite different. If the final destination of a visualization is newsprint, which is generally still a CMYK printing process, the visualization will not only have a limited color gamut, but the contrast is also reduced. If changes in lightness values are used to encode information, it's important to verify that they would hold up on news print.
Designing with the color blind in mind is becoming increasingly popular. In addition to Color Brewer's (which is very cool) functionality, Adobe's Creative Suite 4 includes color profiles that allow users to proof their designs and images so that they can see what they look like to a color blind viewer.
Like Calvin and Ketrina, I am also wondering if there is some guide exploring what different colors mean in different cultures.
