Over at ProLost, Stu Maschwitz presents a nicely illustrated overview of the topic of memory colors. Go and give it a read. This is a fascinating subject that I myself stumbled upon three years ago while researching the IPT color space (long story), and there’s a whole body of academic research in the imaging science world on memory colors going back sixty years if you’re interested in digging deeper.
It’s also a topic I’ve been researching for my next book on color correction, in which I’m planning on citing the sources covered in this blog entry to try and provide a more data-driven framework for discussing why we colorists make the kinds of adjustments we do. Until then, here’s a super-quick overview of articles to provide some food for thought. My apologies for the lack of imagery, I’ve not had the time to get illustration permissions from all the papers I’m citing here (that’ll have to wait for the book).
An excellent starting point is a great article from 2004 that was presented at the IS&T/SID Twelfth Color Imaging Conference (coauthored by Clotilde Boust and too many others to list here) titled “Does an Expert Use Memory Colors to Adjust Images” (the paper is available on PDF here). It cites experiments tracking how a number of expert photoshop artists in controlled environments identify regions of a series of images to isolate for specific correction, and the direction in which the colors were adjusted. Correlating data from their work on four specific images, it was found that the following image regions were consistently isolated for targeted adjustment (secondary color correction, for you film/video folks out there):
- Skin Tone
- Green Grass
- Blue Sky
Experts preferences for these three colors were found to overlap, with all the test subjects adjustments pushing those colors into the same directions when their individual adjustments were plotted with vectors on a u’ v’ graph. It’s an interesting paper, and I’ve absolutely found in my own work that these are the three subjects clients most often want more tweaks made to in any given scene, whether it’s a documentary or narrative program.
What’s really interesting to me is that, while adjustments to a particular subject corresponding to a memory color fall within a particular region, the regions are fairly large, leaving room for individual preference, subject variation, and the influence of lighting (discussed later on). Thus, the data seems to support general guidelines over hard rules.
This paper draws upon information from Sergej N. Yendrikhovskij’s 1998 paper, “Color Reproduction and the Naturalness Constraint” (available from Wiley Interscience). It’s a long and technical document (I freely admit I’m no mathematician) but it’s also filled with a lot of valuable background on the search for a meaningful definition and means of measurement for color image quality in a world of incredibly different display devices and print technologies.
Chapter 2 specifically deals with memory colors. It cites E. Hering as the originator of the term with regards to “the colors that are recalled in association with familiar objects” (quoted from “Outlines of a Theory of the Light Sense,” Harvard University Press, 1964). In a great example for us colorists, Yendrikhovskij describes memory color within the context of a person looking for a banana at a grocery store. Taking for granted the phenomena effecting the eye’s perception of color in that situation, the banana as perceived in the bin is compared with the memory color of that person’s ideal banana (how that ideal memory color is formulated is an entirely different topic).
My key takeaway from this example is a) memory colors have a concrete effect on the appeal of a visual subject to the viewer, and b) someone’s ideal color for a thing may have nothing to do with that thing’s exact, photometrically measured color. This is when the client tells you, “I don’t care if it’s accurate, I want it to be yellower and more saturated!”
Digging deeper, I found a great study by C. J. Bartleson (Eastman Kodak Company), in the January 1960 issue of the Journal of the Optical Society of America, titled “Memory Colors of Familiar Objects” (available from OpticsInfoBase). The goal of that study was to identify, based on fifty observers (with percentages of “technical” and “nontechnical” people alike), what colors were most consistently associated with specific, highly familiar objects (I’m paraphrasing here). This paper found 10 objects for which viewers exhibited consistent preferences across the group (plotted as a close cloud of points on a hue vs. chroma graph) that includes:
- Red Brick
- Green Grass
- Dry Grass
- Blue Sky
- Tan Flesh
- Green Foliage
- Inland Soil
- Beach Sand
This is clearly a longer list of subjects that may elicit audience expectations, but I find while the previous skin/grass/sky subject isolations apply well to requests I routinely get from clients, this longer list applies more accurately to the expanded list of things that I’ve found myself fiddling with in various programs, before the client even makes their first comments. For example, I’ve not had very many clients make specific requests about targeted adjustments to dirt, but having color corrected a feature that took place in the desert, and many scenes in beach environments, I can attest to having spent lots and lots of time obsessing about the ideal colors for earth and sand!
Getting back to audience preferences, the earliest paper I found related to this subject is by J. P. Guilford, from the December 1959 American Journal of Psychology, “A System of Color Preferences,” (available at JSTOR) testing 40 observers (20 men, 20 women) on their general preferences of colors, irrespective of object associations. Color chips were rated from 0=”most unpleasant imaginable” to 10=”most pleasant imaginable.” I’ve not fully digested the entire article, but while the tested subjects make the study very region specific (they all lived in Nebraska), I’ve been curious to see if the paper identifies any preferences I’ve observed in my own client sessions. I’d also be curious to see if an identical study done today would reveal changes in color preference over time (in Nebraska).
One last paper I’ll mention, Scot R. Fernandez and Mark D. Fairchild’s “Observer Preferences and Cultural Differences in Color Reproduction of Scenic Images” (available from the Center for Imaging Science). This is another series of experiments testing observer preferences, but this time sampling an international group of subjects (Chinese, Japanese, European, and American) to try and see if there are consistent regional preferences. The results point to statistically significant preferences in different populations, and I think it’d be really interesting to compare these to anecdotal observations of colorists doing client work within each of these regions. In two examples cited by this paper, Japanese seem to prefer a lighter image compared to the other groups, while Chinese seem to prefer higher contrast images then do Americans and Japanese.
So, is all of this research worth pouring through to a working professional? As another colorist of considerable experience mentioned in a mail list elsewhere, much of the professional colorist’s work is intuitive, based on years of experience grading many programs for lots of different clients. Any colorist who’s been in the business for years is an individual treasure trove of this kind of information.
Personally, I find it comforting to read through a body of research investigating how the audience perceives the image, and that backs up many of the things I do on every project. As a writer and occasional instructor for color correction classes, I also think it’s extremely useful as a starting point for discussing how to begin going about making color adjustments with people who are new to color correction. Lastly, it’s a nice rationale to offer to clients that want a specific adjustment made that one might consider ill-advised.
However, I think an important point to make amidst all this research is that we shouldn’t all be grading skies, grass, and skin tones identically in every project we work on. That would be dull, it ignores individual variation, and it also doesn’t take into consideration the important role that scenic color temperature plays in keeping all elements of the picture unified, rather then ripping an image apart to look like a poorly made composite. The dominant light source in a naturalistic scene affects everything it illuminates, and in a typical grade I think that subject highlights ought to reflect that. In the 2006 paper “Color Enhancement of Digital Images by Experts and Preference Judgments by Observers” from the Journal of Imaging Science and Technology (available here), the authors state:
“The expert also follows some rules: the corrections must be plausible inside each segment and for the whole image, in relation with the illuminant of the scene. The images are accepted by observers in relation with the presence of memory colors and when the treatment of the whole image seems coherent.”
Furthermore, it would be a mistake to interpret this type of research in too literal a fashion, so that visual storytelling is held hostage to audience expectations of the way things ought to be. Instead, whatever the client and you decide to do from a creative perspective, you are at an advantage if you’re aware of how you’re either playing into, or against, the audience’s expectations (whether the audience is conscious of it or not).
That to me is the real fun of the job, finding the right tense scene in that thriller where pushing the color very gently against type, if you will, can give the audience just the right sense of unease. Or conversely, grading the kiss in a romantic comedy knowing with full conviction the audience ideals of hue for skin tone and background sky that will put the shot right over the top.