current page

history

user

Lecture on Oct 12, 2009. (Slides)

Guest lecture by Jason Chuang.

Readings

  • Required

    • Charting color from the eye of the beholder. Landa, Fairchild. (pdf, abstract in html)

    • Chapter 5: Color and Information, In Envisioning Information. Tufte.
  • Optional

    • ColorBrewer: Selecting good color schemes for maps. Cindy Brewer (html)

    • BruceLindbloom: Useful color information, studies, and Files. Bruce Lindbloom. (html)

    • Meet iCam: A Next-Generation Color Appearance Model (pdf, html)

    • Color2Gray: Salience-preserving color removal. Gooch, Olsen, Tumblin, Gooch. AMC Transactions on Graphics. (pdf, html)

    • A framework for transfer color based on the basic color categories. Chang, Saito, Naakajima. (pdf)

Comments

nmarrocc wrote:

Colors have a lot of limitations. Its interesting to think that monitors are not perfect fidelity. If I were an artist concerned with color I might not even want my art to be released online because most monitors would interpret it differently. Its also interesting to think about the limitations of the human mind on interpreting color. If you design a graph that uses color to distinguish between graphical elements how will different people interpret this graph. Will color blind people be able to read it too? Can we ensure that our color choices are visible to everyone? How can we avoid these unintentional optical "illusions." Can we build these color perceptions transformations into cameras to prevent stuff like yellow saturated pictures. Maybe eventually some program that looks at context of surrounding colors and alters colors to make sure that people can distinguish them, or transforms them to make them closer to how a human would perceive them in the real world.

fxchen wrote:

Today's talk was fascinating! I used to be a web designer and had understood how some colors worked better with others, but never understood that colors had a physiological science behind it. The fact human cognition and culture languages define perception makes the topic all the more compelling to study this topic.

Looking at the Lindbloom site, one can find conversions between each of the different colorimetric representations. Some of these concepts were definitely introduced in a graphics class when I was programming in OpenGL. I can't help but wonder how accessible the web as a whole is towards those who are colorblind?

Finally, does anyone have links to the papers describing the studies on ColorBrewer?

bowenli wrote:

Landa, Fairchild: I'm having trouble accessing this, even with a library proxy.

In response to nmarrocc, I think this is kind of an interesting philosophical point. On the one hand, it's true that you can't control color very well, since it depends on so many factors like lighting, medium, etc. But should your design hinge upon the color being exact? I would say no, it shouldn't be necessary to control everything down to that level of detail because you can't. Even if it were lab conditions and everything was the same, people see different. For example, my left and right eye perceive color differently. But knowing that doesn't mean I can't enjoy art or design or read graphs. I think the best art is the kind that can transcend experiences.

Tufte: Laughed at the suggestion that yellow be used for window borders. I think nowadays we use grey, or very thin lines to counter the effects of "1+1=3". There is also the use of 3d effects such as shadowing, etc. I think that is one topic missing from the color discussion. Here color is used as a single blot to completely color some element. But color can also be used to add gradients, shadows, etc.

vad wrote:

I was thinking a lot about the color blind after that lecture. Seeing as we have these semi-complete models of the different colorblindness conditions I was wondering whether there is any software which would enhance a normal display for a specific type of colorblindness (sort-of the reveres of VisCheck). I didn’t find exactly what I had in mind but I did find eyePilot (http://www.colorhelper.com) it’s an interesting application that I am sure helps the colorblind. It has some cool features like being able to highlight specific colors by name, and I think it might even come in handy to the fully sighted person. eyePilot augments the display with some cool color interaction techniques that would fit well natively into visualizations; have a look at the demo or get the 20 day trial.

jcchuang wrote:

Sorry about the link for the Landa+Fairchild paper.

I've updated the link to use the Stanford Library e-Journal instead of the publisher (American Scientist) website. Please let me know if you're still experiencing problems.

jieun5 wrote:

The criticisms on rainbow maps from lecture slides seem to nicely complement the optional reading on Color2Gray. These two contrasting issues (the downsides to using a color-map vs. trying to make-up for the lost color-information when performing a dimension reduction into grayscale) are especially fascinating to me, because I'm interested in generating perception-based spectrograms for audio signals, in which visually salient features of these color-maps correspond to perceptually salient features in the audio.

Given the four points outlined in the lecture slides on why the use of rainbow maps cause worries, I wonder if color does *any* good in the context of rainbow maps. By simply mapping data values to grayscale values (rather than to a sequence of hues), all four "cons" of using the rainbow map would go away: (1) grayscales are less prone to being segmented into classes because of lack of concrete linguistic color-names that divide different shades of gray into firm categories; (2)while hues are not naturally ordered, gray-scale values (white to black) are; and (3 & 4) the problem of lightness emphasizing certain scalar values, or low luminance colors hiding high frequency, would go away when we just have a single dimension (gray-scale values) to work with. Unless there are any significant "pros" for using color that I'm overlooking, I wonder why we don't simply use a gray-scale map on all situations that currently make use of a rainbow map.

In contrast, the idea presented in Color2Gray of preserving saliency information of color when performing a dimensionality reduction to a grayscale map was very insightful. I liked how the principles behind these procedures are all perception-driven; the author's observation for "the growing trend in computer graphics of using change-based mappings for image representation and manipulation" seems consistent with how we tend to pay attention to *changes* in stimulus (as opposed to their absolute values). Finally, computing target differences delta_ij (described in Section 3.2) as a piece-wise function on 'chromatic difference' and 'luminance difference' (rather than a blend of the two) was a surprising algorithm choice, but I guess this makes sense because the two are rather orthogonal perceptual properties (as shown in the Munsell model), and adding them in the same dimension would create a muddled effect.

wchoi25 wrote:

I think the example on p.82 is especially bad for choosing blue as the color that would end up on most of the areas depicted. In cartographic maps, it is obviously important to make an effort to map elements to their natural colors. Using the same blue used for oceans and large bodies of water to color most of the continental U.S. makes the already jarring color scheme that much more unnatural.

At the same time, Tufte's advice to "use colors found in nature" should probably not be mistaken as a call never to use unnatural colors. I think unnatural colors actually work very well when the desired intent is just that - drawing immediate attention even to the point of fading the rest of the supporting visual elements to the background. This reminds me of why the baseball statistic brushing example from lecture was particularly effective. I think the bright yellows really stand out from the mostly black to gray datasets in all panels. Similarly, I think unnaturally bright colors do have their place in visualizations where high contrast and identifiable parts are the goal. I feel that many pie charts benefit from having an unnatural combination of colors just because of the high contrast they provide and a strong color can be easily identified across multiple charts.

jsadural wrote:

Hi Jason, Your pdf slides have no text

malee wrote:

Of all the Tufte reading, the topic of color seems to be the easiest one to understand, and it surprises me that it is so easy to get wrong. Honestly, why would you use blue to designate landmass on a map? Color is pervasive, and most people would probably point this discrepancy out to a child if he were coloring in a map with the land blue and the water green. The use of adjacent jarring colors (for no purpose) also confuses me. I suppose people can choose colors in a rush, but color is sensed so quickly that a quick look at it should warrant an edit. That being said, I do agree with @wchoi25 because the occasional use of unnatural/crazy colors can be particularly powerful in a visualization. This, of course, is in line with the notion that less is more in dataviz.

malee wrote:

Speaking of colors... http://img25.yfrog.com/i/wlq.jpg/

aallison wrote:

Very interesting stuff!

The noisy image of concentric rings on page 82 of Tufte made me think about other interactive color effects that might otherwise go unnoticed if not pointed out. http://www.boingboing.net/2008/07/21/bauhaus-tops.html for example are a series of tops, that when spun, create different, subtle color effects.

I certainly see Tufte's point about rainbow palettes for data graphics. If there is no smooth transition between colors (i.e. if you have discrete colors as labels), then the mind cannot order them properly. Of course, smooth gradients of colors make it harder to distinguish absolute values from each other. Tufte's point about using elevation lines to mark divisions is a good one.

What tools exist to help developers and visual designers create such lines from large sets of discrete values?

vagrant wrote:

I am an ardent follower of the leaning that solid color is generally a more suitable alternative to patterns for information visualization. Noise is one of the most irritating vices of any visual presentation, and favoring simplicity and easy perspective, I have a particular dislike for gridlines and stripes. I found Tufte's chapter vindicating. I also appreciate the rule that color should be quiet and subtle in most places, allowing for emphasis via brighter, bolder colors on a sparing basis.

I wish there was more exploration in the readings on the topic of hues. As an owner of multiple high definition displays, I find that the average display of the modern day--the flat panel LCD--to be rather paltry in quality when it comes to displaying shades, and to be inconsistent from different viewing angles (e.g. from the side of above/below). It would be nice to have some idea of how representation design might compensate for or respond to the deficiencies in modern digital displays.

rnarayan wrote:

aallison - while I agree with your comment and Tufte that redundancy and partial overlap resolves ambiguity with regard to the graduated scale of colors with contours (elevation lines) on the Bathymetric chart, one drawback is that the use of two graphical signals such as line and color, render them unusable for the vizualization of other info in the same image. And typically, maps of weather, terrain, land use, etc. requires simultaneous display of multiple variables such as temperature, precipitation, etc. Instead, another technique uses contours with parabolic ridges (http://www.win.tue.nl/~vanwijk/ecm.pdf). The smoothly varying bands between the discontinuities are easier to track (the transitions are actually a result of an optical illusion) and has the advantage of leaving color and lines for other purposes (per the paper). Shown below is a weather map using above.

Another issue with using color gradients is the perceptibility of small increments when the data series is large (e.g. when color is used to encode non-geospatial data such as time). I ran across an example of such use in this paper (http://cs.swansea.ac.uk/~csed/papers/VisualisationOfSensorDataFromAnimalMovement.pdf). Although small gradients in color should be avoided in the first place bcoz of gamut mapping, etc. (as well covered in class).

mikelind wrote:

I've found this subject incredibly interesting to think about, especially coming from a design perspective. I also hadn't thought much about color blindness until recently, when I was driving with a project partner to home depot to get some supplies, where he ran through two stop signs and almost ran a stoplight before I mentioned it to him. I learned that he had colorblindness which prevented the lights and signs from really standing out, so it got me thinking a lot about how we can use color and how necessary it is to encode some redundancy into our visualizations and designs that focus on color. It seems that color can be incredibly effective if done carefully, but it may require more thought to produce something that can deal with all of the potential problems that color can introduce for different people. I think vad has a really interesting idea regarding altering displays for color blind people. I'll have to look into eyePilot a bit myself.

alai24 wrote:

@nmarrocc I think having the colors of a work being distorted is pretty much unavoidable even in the analog world. Lighting and weathering could have a bigger impact on how the colors look than a monitor's configuration.

I found Tufte's suggestion to use colors found in nature to be pretty insightful. Culture might have a lot to do with what color schemes appear eye-pleasing to a person, but pretty much everyone is exposed to nature's palette.

dmac wrote:

After reading Tufte and listening to lecture I started realizing how many web sites use low-saturation colors. Google, for one, often uses unsaturated colors as decoration or structure so as not to distract from data. Their Charts API uses a particularly nice set of colors for visualizations.

jqle09 wrote:

The ecm.pdf paper rnarayan cited had some really nifty examples using contours with colors to display precipitation and temperature. I perceived contour lines as displaying height and so these particularly popped out displaying the precipitation very clearly. It was really easy to see dips and mountains of rainfall, more so than when color is used to display precipitation. Really cool stuff.

This topic is really interesting and has looking more closely at the colors I picked for my emacs color theme. For me picking colors for text has been quite challenging as I want certain text to pop out more, while everything must still remain particularly readable on a single background. I also want it to look pretty. It's nice that I get to experiment on myself, figuring out how I perceive colors, and the ideas from this lecture have helped me approach this. Although monitor to monitor the display of my text gets changed and is quite troublesome. Also changing fonts seem to also change how the colors are perceived (or rather how the font is perceived). Some colors seem to make characters feel thicker/fatter so they look nice and round, and as mentioned in lecture this really depends a lot on the background as well. But knowing all this should help us choose colors for visualizations as well.

akothari wrote:

Last week was truly a color week. first, jason's talk on monday, followed up with the hci talk on colors by Stephen Palmer (UCB). I always thought it was more art than science. But the talks definitely persuaded me that there is aesthetic science of colors. I was personally very intrigued by how colors are perceived differently in various cultures/countries. For example, Stephen noted the perception of different colors in US and Japan, with the latter preferring lighter colors. It would be interesting to see how different quantities of color affect perception. For example, purple is a great color - but too much of purple can make it look very gaudy and ugly.

joeld wrote:

Of all the variables we've discussed in class color is probably the one that evokes the most direct emotional response. For example, the low-saturation palettes used in the Google Charts API are a nice conservative choice because they are "soothing". In particular, I might argue that the presence of gray in a set of otherwise incompatible colors reduces the amount that different colors clash. This technique can be used to devastating effect at manipulating attention.

I spoke to an interior designer once about her technique and she referred to a trick in color schemes called the "eye of the dragon". The basic idea was to have most of the space in a subdued tone - matching browns or other neutral tones - in essence reducing the scene to a colorized black and white image. Then she would introduce one or two pieces with a very striking red or blue. Because this is the only element of that color present in the scene the eye is drawn to it very strongly.

I'm also interested in the notion that certain colors have subconscious emotional associations, but perhaps that will have to wait for another post.

saip wrote:

Color, besides being vital in conveying information, plays an important role in design and aesthetics. I'm a huge fan of Colourlovers (http://www.colourlovers.com/), an amazing site with nifty color design and contains a large library of color palletes that one can use while picking color combinations for visual work.

I just finished reading Chang et al's reading on color transfer and was amazed by the possibilities it holds for data visualization. It presents an amazing way to transfer (approximately) the color palette from one image to another. I can imagine a number of applications of this to things that I regularly do. For example, it will be useful to select an item in my photo collection with good color and copy/paste its colors on other images with poorer color quality. This can be particularly useful when shooting a panaromic picture, where using a medium quality camera in the Auto mode, we often end up getting slightly different colors in the picture depending on the direction of the source of light.

rajsat wrote:

Speaking of the use of color to display effective visualizations, I was interested in the use of color to aid stereoscopic viewing- about how 3D viewing works. Though its a little off-topic, Its interesting to note how red/green or red/blue 3D glasses work. Basically there are two images on the screen- one in red and the other in blue(or green) and the filters on the glasses allow only one image to enter each eye. Our all-powerful brain does the rest. But, this technology is outdated now and they use have switched to polarized lenses. Another application of color is in Autostereograms- the only difference between this are normal stereograms being that we do not require glasses to view these 3D images. The image shows 2D images of the same object from slightly different angles to the left and right eye, thus allowing the brain to reconstruct the image in 3D. But this works only at a certain viewing angle. Here are some wonderful examples.

nornaun wrote:

@akothari. I think the topic of colors in cultural context is interesting too. I believe that different cultures have subtle differences in interpreting colors. I grew up in Thailand where every kindergarten students are taught to map colors with days of a week. (FYI, mon = yellow, tue = pink, wed = green, thr = orange, fri = blue, sat = purple, sun = red. Read more at http://en.wikipedia.org/wiki/Color_of_the_day) For me, wed is so green and I was surprised that this is not an international notation. Colors are also used to symbolize ideas or entity differently. For example, in Thailand, purple is color for gays while it is rainbow in the US. I would say that the assignment of a symbol to a color is quite random. However, the way people from a certain background interpret the colors is not.

zdevito wrote:

The castle optical illusion points out an additional limitation of color in addition to showing the opponent processing of red-green and yellow-blue. If you look at the opposite-color castle picture for a long time and then switch to a blank screen, instead of seeing the correctly colored image, you might see a blob of blue/green colors that roughly approximates the original image, but is nowhere near as clear as the image seen when the colors are perceived in the black/white image. It seems like the visual system has much lower resolution for color determination than for other features such as lines or changes in brightness. This realization has been put to use in image/video encoding where the color values for a pixel are frequently changed into luminance + 2 color values, and the color values are down sampled by up to a factor of 2 without much perceived loss in the original image.

tessaro wrote:

Olver Byrne's Euclid (scanned edition here: http://www.sunsite.ubc.ca/DigitalMathArchive/Euclid/ ) is such a singular work, that it raises many questions about encoding of information beyond the discussion of color. It ambitious integration of text, size, orientation and symbol as well as color is itself a fascinating study in the overlapping congruencies and conflicts of differing strategies we employ when parsing information.

Tufte quotes Paul Valery's axiom "To see is to forget the name of the thing one sees" later in chapter. With respect to Byrne's treatment of Euclid, one is tempted to ask what exactly what is going on in the mind of the 'reader' of the proof's text. I find myself jumping from the logic of one system to the other, jumping from judgements of dashed-lines to hyphens, from scale dependent notation to scale invariant geometry. (It would be interesting to use eye tracking to build a composite image that traces the glance of a reader of one of these pages. What areas of the brain would light up if we viewed the eye tracking data with a parallel fMRI?)

The almost punning nature of this composite challenges the mind to float between exposition and diagram, almost like a rebus. One of the more interesting aspects of this conflation of systems is what it does to the experience of serial thinking in the build-up of the proof, the march of logic toward Q.E.D. and the parallel thinking unique to visual processing and reflection. The self evident truth of the many visual proofs of the pythagorean theorem (like the chinese example in the margins) runs in almost direct opposition to the unwieldy logic of Euclids proof. In that sense, Byrne's technique may be doing another trick; that of revealing the obtuse construction of the proof by way of a graphic taxonomy of its verbose complexity. Mondrian is thus turned into Proust.

cwcw wrote:

As part of a science unit on color & light, I used to teach my 5th graders about the physiology of how we see colors. This was, of course, on a much more basic level than Jason Chuang's lecture. I wish I'd been able to attend his talk before teaching my students!

Still, a lot of what he talked about was familiar to me--about primary colors, and about cone/brain fatigue that causes one to see colors that are in fact not there (I used the flag on this page with my students http://www.triangleparkcreative.com/tips/print/color), and about sensitivity to people with color blindness. I would love to go back and share the castle example with them--that was really cool.

As another exercise, I would have my students look around their rooms before they went to sleep, to recognize their rods worked well in dim light (they could still see shapes), but their cones did not (it's hard to tell the color of things in the dark--they appear much more grayscale to the human eye).

I would usually put the daily schedule on the board using multiple colors--not so much as a code to convey information (as much of our work in CS448 deals with), but more because it was eye-catching and (if still done properly, according to certain principles Jason discussed in his lecture, like use only a few colors, and strive for harmony) visually appealing.

One other item I thought was really fascinating, was that most cultures name colors in this order: black & white --> red --> green & yellow --> blue --> brown --> pink & purple & orange, etc.

gankit wrote:

This discussion is really interesting. I had never thought of color in a cultural context. Now that I think about it, different interpretations of color in different cultures makes a huge difference in how someone percieves your visualization. It is almost equivalent to giving misleading information if you get the cultural context wrong.

This got me thinking on what emotional response does color evoke. And then, I found http://www.color-wheel-pro.com/color-meaning.html

It seems that this has been well researched and there are some colors that provide a reliable estimate of the emotional response. I feel that understanding this can aid strongly in deciding which colors to use to highlight the different parts of the visualization. Also, when encoding ordinal variables, we can choose colors accurately based on the emotion we want them to represent.

cabryant wrote:

Oops. I just discovered that I posted my comment for this subject in the section for September 23. Here is a transcription:

Of the four uses of color enumerated by Tufte, the nominative, the decorative, and the imitative carry the most promise. In fact, the only quantitative renditions of color that seem reasonable are those that seek to imitate reality (e.g. visual correlates of depth, temperature, and the like). Those that do not, run the risk of unnecessarily violating Tufte's maxim: Above all, do no harm.

One color effect that deserves additional consideration is that of desirable instances of 1 + 1 = 3. The road map on page 93 of Envisioning Information makes the case for color use that blends the boundaries between decoration and nominality. Although this aspect of color would be incredibly difficult to automate (indeed, it would likely be foolhardy to do so), there may be merit in providing support for such manual alterations in visualization software.

Finally, the description of the use of Munsell's coloring scheme to systematize the world of manufacturing begs the question: although Munsell drew upon nature as the source of inspiration for his schemes, has this application altered our collective perception of color, subtly affecting and/or standardizing our preferences?

codeb87 wrote:

One of the most fascinating parts of Jason's lecture was the linguistic study of many cultures' terms for colors as the language and people evolved. The fact that this history of evolution was similar among cultures, and corresponded so closely with the priorities of our visual system, was a compelling argument for just how well defined our human encodings are for colors.

I was also impressed by Munsell's organization of visible colors, but I find it limited in its discretization. I think that a continuous, mathematical representation of the color space would open up a host of applications to both designers of visualizations and artists. Particularly I am interested in the possibility of mapping a continuous function into the color space. By choosing simple functions or parameterized exotic ones, and being able to map these functions to certain regions of the color space, I think there are endless possibilities for creating striking visual color transitions. Although this wouldn't be appropriate for a simple data visualization, I think it would open up a whole new realm in which we as programmers could play with the human visual system, with striking artistic effect.

Leave a comment