Color

From CS294-10 Visualization Fa07

Jump to: navigation, search

Lecture on Oct 22, 2007

Slides

Contents

[edit] 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

[edit] Mark Howison - Oct 16, 2007 07:39:12 pm

One of the rules of thumb that Brewer gives for working with CMYK color is:

"5. Equal percentage steps don’t look like equal visual steps; use bigger steps in higher percentages (Example: 5 to 15 looks more different than 80 to 90)"

This seems to agree with both the Weber-Fechner law (that perceived differences requires greater differences in stimulus as the intensity of the stimulus increases) and Stevens' power law (that perceived intensity is a power function of stimulus intensity), as long as the exponent is > 1. In fact, Wikipedia claims that the empirically measured power law exponents for lightness and saturation are 1.2 and 1.7 respectively. Those values also agree with Brewer's claim that, "There are also usually few perceivable steps available in saturation differences," since the exponent for saturation is larger, and therefore saturation perception is more non-linear.

[edit] Robin Held - Oct 21, 2007 05:30:30 pm

Brewer makes a good point point when she states that the functionality of a color map should come before its aesthetics. If one worries too much about creating a "pretty display," one could easily forget to include a set of colors that is both effective and useful by all potential readers--colorblind or not. Plus, as Brewer states, it is impossible to choose a color scheme that EVERYONE will like. Of course, usefulness and pleasing aesthetics aren't mutually exclusive--oftentimes there will be a variety of effective color schemes available to a designer and at that point, artistic license should be exercised.

[edit] Willettw - Oct 22, 2007 12:40:14 pm

A few more quick thoughts on remapping the color space in images - particularly in agricultural surveys - as mentioned in class. A lot of the aerial and satellite imagery you see of cropland is colored in predominantly red hues. Image:Infrared1.jpg

Typically this is because crop health images have been shot in the near-infrared using false color film. This is because most plants absorb heavily in the infrared (and the lower part of the visible spectrum, hence an abundance of plants that appear green to the eye rather than red) while soil has a very different absorption profile, meaning that differences in the density and health of the crop can be picked up much more easily in an infrared image. The red color is actually an artifact of the false-color film that was traditionally used to take these sorts of shots and digital infrared images usually mirror this color output using a red-green-infrared CCD (whether the color choices for the film were perceptually motivated, I don't know).

I spent several summers as an undergrad taking these sorts of infrared aerial photos for an agricultural consulting firm. Typically we'd shoot using a digital infrared camera and make prints of the infrared images. We'd also separate the channels of the image and calculate a more detailed "crop health index" based on them which we'd then map to a rainbow scale (we used Matlab to do the processing and probably did choose this because it was the default).

Image:Infrared2.jpgImage:CropHealth1.jpg

In retrospect, that color mapping probably wasn't the most effective visualization and a gradient mapping that varies saturation and/or brightness based on the output of the crop health algorithm is probably more appropriate. It strikes me as a little strange that in the false-color infrared images, the healthiest plants actually end up being the most red - a bit of a counter-intuitive mapping. The crop health images try to counter this by making green the top of the scale - unfortunately this results in a scale that goes from blue (worst) to red (bad) to yellow (better) to green (best), a very counter-intuitive reading of the scale since we typically expect blue to be at the opposite end of the scale from red.

[edit] Kenrick Kin - Oct 22, 2007 02:06:20 pm

During lecture something I found to be an interesting point about color is that the user should be able to name the different colors in a visualization. It sounds trivial and it seems like something we take for granted or don't really think too much about, but it does matter. Being able to verbally identify a color is important to effectively communicate the properties of a visualization when discussing it with another person.

[edit] Kenghao Chang - Oct 24, 2007 09:07:10 pm

Tufte mentioned several coloring principles in Envisioning Information. For example, background color should be grey so that it doesn't compete with colors that carry info, prominent colors such as red should be spread out the image so that i won't clutter the page, and so on. This makes me thinking that there are many layout principles as well in graphic design.

In addition, from the class of design principles, we learned that we can use optimization techniques along with several design constraints to get the best result. Combining those two, this makes me think whether people have turn those design guidelines in coloring and graphic design into numerical constraints? If yes, have people tried to use optimization methods to come out best graphic design?

[edit] Mcd - Oct 28, 2007 01:05:35 pm

Brewer's piece was very informative. I looked around earlier in the semester for some help on color space translation, and found a few sources that helped me understand, but none as well as this. Her discussion of the inherent lightness differences in various hues (yellow is lighter, e.g.) was particularly helpful. I then wondered if this might have something to do with the overlap in the red and green curves (where yellow lies) in the cone response diagram that Maneesh showed. Is yellow's perceived lightness the result of this doubled cone response?

[edit] James Andrews - Oct 29, 2007 12:48:53 am

Tufte's argument for the use of color in a proof of Euclid seems rather weak -- looking at the proof itself, the color information is at times quite painful to follow, and only for non-edges saves me any back-and-forth between proof and picture. In the version that eschews text labeling altogether, the notation seems ambiguous in places; lines like "also draw ----- and ------" where the two -----s are identical are disconcerting. When he adds text to the color version, it seems he has to shrink the text down quite a bit to fit it, and even though his diagram is still larger than the non-color version it is harder to make out labels on it. The shrunken size also emphasizes where the color coding breaks down, as edge CA seems difficult to find when one tries to use the color instead of the text to find it. Meanwhile, the fact that he has inadvertently covered part of the CH line with red implies a lack of attention to how well the encoding really works for a proof. Even without considering printing cost, I'm left unconvinced that coloring the proof was a good idea.

In general, the problems with coloring Euclid's proof seem like they hinge on scale -- as the complexity of the proof goes down, color is great; the ambiguity would disappear if only a few labels were needed, and the text size issue would also be a non-issue as there would be no need to cram for space. But as complexity goes up, color would likely be completely unreadable; label ambiguity and size issues would be terrible. In the context of a textbook which plans to cover much geometry beyond Pythagoras's theorem, desire for a consistent style would likely rule out any such coloring approach.

It does seem like the real win from using color in geometric proofs would be in adding color guided by interactivity, so clicking parts of the proof would highlight their corresponding elements in the diagram. By adding and/or highlighting detail incrementally, the scalability issues could be solved and maximum benefits from the coloring could be achieved.

[edit] David Jacobs - Oct 29, 2007 10:34:20 am

One of the things I've always found interesting about color perception is the seemingly arbitrary mapping of the complete color space (infinite in dimension) to a 3D space. This is great for display manufacturers, because now we can create projectors, LCDs, TVs, etc. that only have three kinds of phosphors that can be combined linearly to display metamers for nearly any true color. What I'm curious about, is how these displays look to people sensative to more colors of light. Some women have a condition known as tetrachromatism, a condition in which the retina contains 4 types of cones rather than 3. For them, the whole science of the CIE color space doesn't really apply, so are all the modern display devices awkward and weird looking compared to the real world, or is it just an inconsistency across different display devices?

[edit] Amanda Alvarez - Oct 29, 2007 02:25:46 pm

I really liked Brewer's paper, especially the graph of color schemes and examples which really help to illustrate the mapping of color to data structure.

David: I don't think tetrachromats have any augmented color perception compared to trichromats, despite the fact that they have four cone types. As per the Wikipedia article: "It is not known how these nerves would respond to a new color channel, if they could handle it separately or would just lump it in with an existing channel." The thought is that there is no way for the new cone channel to fit in with the existing color-opponent information pathways (red-green, blue-yellow), so it is just lumped in somewhere and the information is lost. Obviously a lot of work still needs to be done to assess the color vision capabilites of tetrachromats, but I feel like their extra cone does not add another axis to the CIE space; maybe they just have heightened sensitivity, or can pick out finer gradations? In either case, one would think that displays might look 'weird' even to normal trichromatic observers, considering they have a reduced gamut compared to CIE color space, which is based on a 'standard observer'.

[edit] Jimmy - Oct 29, 2007 04:37:04 pm

Tufte gave an example on encoding colors into quantitative information. In the ocean map of Japan Sea, the ocean map was colored according to its depth and height. The use of numbered contour lines can help on the accuracy of reading. The color itself might be inaccurate because the reading of colors can be disturbed by contextual effects, such as edge fluting and simultaneous contrast. So adding the lines could help on fine value distinction and improve the scale precision.

The human perception on colors is surprisingly sensitive to interactive contextual effects. As show in the lecture, the same color can look very different on different backgrounds. In spite of the reading convenience of colors, they could be misleading and not able to delineate what’s obvious. The boundary lines for the color map are shown to offset these distracting effects.

[edit] N8agrin - Nov 03, 2007 10:43:58 pm

Brewer's article helped me understand a bit more of the science behind color and some of the better guidelines to follow when using color to encode quantitative or qualitative data. I agree with Amanda, Brewer's color mapping is very useful for demonstrating her various schemes of color mapping based on data.

I also found Tufte's suggestions on how to use color on page 81 to be informative. He lists the uses as to label, measure, represent or imitate reality, and to enliven or decorate. Of course he throws in for good measure his own principle: Above all do no harm. But as usual with Tufte, his principles seem sound, and then he shows seemingly bizarre examples to illustrate his points. The map on page 95, which supposedly is a better version of the one on the previous page is still confusing to me. At once he claims that the map on page 94 is bad because the use of color is uninformative, and then applauds a map whose exterior outline is also swathed in color. I spent nearly a minute trying to understand what the green and brown lines in the upper left of the map were before I realized that the region of the map was more land, and those lines were roads and rivers. To my eye I was now seeing an island, not a province within a larger country. Tufte, how is this possibly better?

[edit] Hazel Onsrud - Nov 05, 2007 09:42:54 pm

As we progress through this course it seems easier and easier to cut Tufte apart...and the visual tools that we have been exploring on the internet, such as the graph of color schemes and more interactive, easily adaptable displays often seem to be able to better illustrate Tufte's general principals and allow us personally to explore them furtur...in all of these respects I agree with the comments above.

I found this lecture to be especially interesting because color seems to be one of the most subjective aspects of visualization, despite the plethora of guidelines available. I contend this because although the use of space, and font and general organization of data is critical to a visualizations success, color is a component of the design that each individual sees differently, and none of us can be sure of what the other sees. Additionally, as mentioned above many people are color blind and have unique conditions that further complicate this component of design.



[add comment]
Personal tools