(due Tuesday, March 10)
Color is a big enough topic that we’ll probably want to spend more than 1 day on it. I’m planning at least 2. For the first color discussion, we’ll have two readings: one on the use of color, the other on some more technical issues.
Chapter 4 of Colin Ware’s Visual Thinking for Design (we’re working through it in order)
Representing Colors as Three Numbers by Maureen Stone. This appeared in IEEE Computer Graphics and Applications, and is a nice summary of the science of color (much better than the chapter of the 559 textbook).
As usual, post a comment indicating that you’ve read these. One thing to think about: how do the technical issues (described by Stone) connect to the design issues (described by Ware).
{ 14 comments }
I was a little confused by Stone’s paper. It seemed to first say that color spaces are linear and then that they’re not. It makes more sense that they wouldn’t be linear and that any linear interpretations is just simplification. Having a linear color system does make computer graphics simpler though as you can linearly interpolate two colors.
I was also a little confused by the method of determining the weights for each color channel. It said to multiply each function by the spectrum and integrate. I’ll have to try it a couple of times to see if I get it.
I understand RGB pretty well and it seems like there’s a defined mapping to XYZ and LMS so I think I’ll be all right.
Chapter 4 in Ware’s book was great. The principles he set forth made sense and seem simple enough to quickly integrate into any visualization.
The first point that I noticed was that the luminance channel is really important. I’d say it was responsible for about half of the channel properties listed. Also, a lot of the examples in the book seemed to demonstrate this concept.
I also noticed that he proposed doing exactly what we did for the design challenge which was to map a zero value to a neutral red-green color and then saturate to show positive and negative values. We saturated with green to show positive values and red to show negative values.
Taking into account the information presented by Ware and Stone, one might easily fall into the trap of generating colors for a visualization based on a set of rules or algorithms. For example, if visualization calls for the classification of a large number of variables, using one of the methods that Stone proposes could produce colors with enough perceivable variance for a user to distinguish the different categories within the classification. I believe that an automated approach would likely yield undesirable results, especially given that Ware suggests a limit of 12 different color categories (which I would personally limit to 9). Shouldn’t the colors be chosen in advance given these existing limits?
Unclassed data present some challenges that could be addressed through a combination of the technical approaches from Stone and the aesthetic suggestions given by Ware. Merely adjusting the lightness on a continuous unclassed visualization, such as a digital terrain model, may limit the perceivable differences, especially with near-white subtle variations. A Munsell system would certainly work if saturation levels were high enough and the data were classed, but perhaps another approach would be necessary to ensure that general audiences could detect changes in an unclassed visualization.
I read, understood and enjoyed Ware’s chapter on color. Here are a few of my findings.
Interesting factoid that rods are wasted on us because of our artificially lit indoor world. Could this lead to evolutionary change eventually?
We are bad at seeing short wavelengths, which kinda makes sense, because short wavelengths don’t travel very far.
The brain is more sensitive to differences in colors than in absolute values of color reflected off of an object’s surface.
The luminance channel carries more information than chromatic channel. Is this why we like black and white photos more than color photos?
The most important single principle in the use of color is that whenever detailed information is to be shown, luminance contrast is necessary. Finer the detail, more luminance contrast is required to distinguish features. The larger the features, luminance becomes less important, and color differences become significant.
Note: Interesting to see Ware use “PowerPoint” as a common noun for presentation slides.
Colors are changed in appearance by adjacent colors. This is called simultaneous lightness or chromatic conceits
Interesting factoid that rods are wasted on us because of our artificially lit indoor world. Could this lead to evolutionary change eventually?
I was not quite taken by Stone’s paper. Partly, this was because of my inability to clearly grasp the linear algebra, which, in turn deadened me to the relevance of the math of color to coming up with a good visualization.
Chapter 4 of Colin Ware’s textbook develops approach to use color to design, based on theory. Cone receptors, the important receptors in human eyes, can accept three subtypes of sensitive. This makes the color vision to be fundamentally three-dimensional. On the other hand, the color appearance, which only takes three lights (red, green and blue), has more than three dimensions. The reason for that is spatial effect, since patch of color is not isolated.
I also find that the channel properties in this chapter are very useful to design visualization. For example, it mentions luminance channel has greater capacity to convey detailed information than chromatic channels. My intuition was the opposite. In my research field, I think we emphasize on color more than luminance, but luminance is useful especially for continuous variables.
Stone’s paper, Representing Colors as Three Numbers, introduces RGB and XYZ colors with detail. There exists a transformation function between the two three-dimensional color systems. In order to properly use the transformation matrix, the RGB values must be linear and in the nominal range [0.0, 1.0]. Different RGBs are defined relative to different reference white.
Using RGB/XYZ creates a quantitative foundation for manipulating these numbers with respect to physical specifications of color in display. Moreover, the quantitative feature makes it possible to model surfaces and lights. It can further helps optimizing human color perception.
I really enjoyed both readings.
As an exercise in my undergraduate graphics class, we went online to a rgb triangle to see how much of each was needed to create what appeared to be white on the screen, which unsurprisingly, required a lot less green than other colors, which does make sense evolutionarily.
Having studied color from a theatre lighting design viewpoint, it’s interesting to see it in a different way. When doing theatre, we focused primarily on what color combinations did and how combining colors in light differ from combining color in paint. In these readings, the focus seems to be in comparing between two colors themselves and how luminosity and other aspects affect those differences. This makes sense, as most visualizations will likely involve comparisons.
Given our discussion last week about gamma correction, I found Ware’s discussion of CIE calibrate monitors and hue perception rather intriguing. One question that the section did not seem to address, however is how can a monitor be considered “properly calibrated” when different “proper” colors can not even perceptually be defined? The instance of color I am referencing is the fact that there are essentially two “true” greens, which occur at perceptually different wavelengths. It raises the question as to whether the true correctness of a color display indeed lies in the eye of the beholder.
Across both readings, however, this idea of negative light appears. Both readings contain a great deal of information as far as the mathematical foundations of current color systems; however, all of these systems seem to operate on the fundamental principle that negative color is attainable. This fact is actually very rapidly disproven in Ware’s book, suggesting that the very mathematical frameworks that the aforementioned uniform color spaces lie in may be a mathematic impossibility. Thus, the question remains, which is more relevant for color analysis: mathematical frameworks or perceptual experimentation.
All uncertainty aside, however, Stone continually stresses the importance of the linear and nonlinear natures of color computation with respect to color matching and selection. Not only do these properties appear important to the accuracy of the calculations, they also bring to the forefront of the color argument which color space is more effective: that based on a linear system or that of a nonlinear perceptual system. After completing both readings, I would argue that perceptual definitions of color are more significant with regard to color selection. While uniform color spaces play nicely in the theoretical realm, it feels as if perception tends to avoid such logical simplicities. For instance, the fact that a white-black color ramp essentially only has 4 perceptually significant encodings, despite covering a maximal distance in uniform color systems, seems to imply that it is what we see which is the more important consideration in color selection that what seems computationally correct. Given that visualizations are designed to exploit the efficiency of the perceptual system, it would only make sense that colors be selected to play off of this theme.
Ware’s chapter on color provides a lot of implications for setting up design principles, especially based on contrasts. Based on perception theories such as the color opponent theory, he goes into great detail on how features such as luminance contrasts are crucial to make specific details visible and salient. Since for most data visualizations the purpose is to make specific significant values stand out from the others, luminance could be the single most important thing to consider for the narrow area of crucial datapoints(fovea, anyone?) while using other color features for less necessary areas.
I also liked the rather short part where he talks about the semantics of color(for example, in Korea it is considered rude – almost a curse – to write other people’s name in red!). It reminded me also of the color chapter of the book ‘Understanding Comics’, where the author says that choosing a specific color palette alone can even link to specific superhero characters. In this way, the semantics of color could be useful in doing a more rhetorical storytelling.
Stone’s paper explains that our tri-color vision is non-linear while the RGB coding system is. So she proposes a new formular to make it correspond better and reduce color perception errors. I had a hard time to grasp the concepts within and am had an even harder time to draw implications for designing data visualizations. Maybe it connects to choosing the right color palette to maximize luminance contrasts.
I’m also quite confused by Stone’s paper. Basically, every transformation in RGB system need to be going through XYZ system. What is the necessarity for this? What is the exact different between the linear and non linear RGB system? How did they change to each other? I understand this is really a hard topic, but the answer is still too ambiguous to me.
Stones paper is really a good compensation for Colin Ware’s Chapter 4. I really enjoyed Chapter 4 since I can understand most part of it. The “Watchband” style diagram of hue saturation (darker and lighter means less saturation) and the difference between luminance and chromatic channels are really new to me, and provide useful design criterias for my future design. The shape from shading and contrast part also gives great thought, but I think these can be used as design methods equally important to the color coding.
Human perception of color is heavily influenced by context. The notion that there is no sense of absolute color, along with the limitations stemming from that fact, are well-illustrated in Ware’s chapter on color. However, Stone mentions in her tutorial the von Kriese model for adaptation, which defines in LMS space some threshold value that we call “white”, then defines all other colors as some ratio of white. This model seems like a pretty handy mathematical tool to help effectively use color in a design.
Ware claims that “the three-dimensional nature of color space is a … consequence of having three cone types in the retina”, then writes in the next paragraph that color vision is three dimensional because “it only takes a combination of three lights to create a full range of colors.” The relationship between these two causes isn’t obvious until reading about the CIE’s color matching experiments, and that cone response functions are linear mappings of color-matching functions.
OK. I’ve gotta admit: I’m a little confused by the concept of XYZ space. I’ll give the Stone reading another go today now that I’m more present — but that was a (sadly, rather fundamental) point of confusion for me.
Another thing that had me a little confused was the Purple Line — mentioned both in the old Ware and Stone. Why does our color space have a straight line between R and B? Is it just that there needs to be a straight line between two primaries in that style of representation of color space, or is there something special about Red and Blue?
The rest of the readings were neat and straightforward — it’s nice to finally see why blue text on black is so hard to read; also, the illustrations of contrasts only in hue are delightfully frustrating to read, reminding me of many a bad powerpoint presentation.
Also: I wish I could see in 12-color Chicken Vision.
Chapter 4: Color
*****************
Perhaps the most important thing that I learned from this chapter was the importance of “Luminance Channel” in understanding motion, stereoscopic depth, and shapes from shading. Also, whenever detailed information is to be shown, luminance contrast
is important which is critical for small text.
One important suggestion that this chapter makes about presenting zero in data analysis. In order to make zero intuitive, a neutral value of red-green and yellow-blue channel is a good choice.
This chapter has shown good pictures to convey the fact that colors can totally destroy our ability to see the shape of the surface.
It is interesting that many great photographers, movie makes and artists still favor black and white over other colors.
Representing Colors as Three Numbers:
************************************
This paper was very difficult to read and extract useful information. I failed to understand the objectives and conclusions of this paper.
From Colins’ Chapter 4: The effective application of color in design is based on understanding of properties of color channels. The Ware’s chapter iterates through important properties of color channels followed by design principles based on them. The perception of color appearance for a patch is influenced by surrounding colors. This implies that in a complex visualization we need to be careful what colors we use for backgrounds and other symbols. Similarly, pop-out effect of different colors depends on this principle. The texture of shape or motion is best conveyed by luminous channel, whereas, chromatic channels can suppers the same information. Additionally, there are also limitation on how many colors we can use together in a design. Some limitation related to choice of color of text that can be adopted while offering the required contrast. This is also critical as many times we need to annotate our visualization with some text information.
Although, many of these color principle are helpful in all sorts of visualization, I was particularly thinking about interactive plots with large number of variables. The emphasis and de-emphasis properties of color (e.g. page 78) can be interactively manipulated to help with visual queries.
From Stone: I haven’t fully grasp some of the concepts, related to generation of color , presented in this paper. I am still little confuse about what the author meant by multiple spectra. I guess I never studied color in this detail before. I will make another attempt before the class to understand the paper.
Ware: It was good to see how the different use of luminance, lightness, saturation, and hues makes a difference in perceiving details, highlights, contrasts, and surfaces. One of the interesting thing is that gray-scale gradients are perceived as shading differences while smooth graduations of color are perceived as category of regions into red, yellow, green, and blue. Also, it was interesting that gray-color is most efficient in showing surfaces, which I think I learned in black and white photography.
I’m surprised at how much linearity is involved in color perception:
– Combining color spectra in light is linear
– Cones have constant sensitivity curves in stable environments
– After projecting color spectra into the space formed by the three sensitivity curves, and assuming a stable environment, cones are linear.
I found it interesting that:
– Producing colors with spectrum identical to the LMS sensitivity curves is required for a display device to create all perceivable colors.
Something I couldn’t figure out:
– In Stone, simultaneous contrast occurs because of differing white among cones.
– In Ware, simultaneous contrast occurs in or after V1.
– If Stone is correct, then cones from different regions of the retina have different whites. (Otherwise, the contrast illusion cannot appear.) This is interesting because cones also have a global white – if a bright point light occurs in darkness, the entire eye is desensitized.