Can You Pass the Online Color Vision Test?

[zoobyshoe]

Scanning at her website again, yeah, there do seem to be some implications of that. Strange, I never noticed that the first few visits.

• "Concetta paints with 100 million colors"
• "See what you have never seen before"

Seems a little misleading when put in the context of her tetrachomacy, which is also a predominant part of her website.

As you say, if she chose her pigments carefully such that a given painting looked color-realistic to her view of the real world image, then it would look color realistic to me (a trichomat) as it would to a color blind person (going the other direction isn't the case though).

If one did make a computer monitor and video card system that can tap into the 4 primary, tetrachrmatic colors properly, and cameras were created with these 4 color channels, and images were created with these four color channels (which would require a revision to image file formats like png, jpeg, gif, etc.), then that 4 color system would work just fine for a trichromat like me, as it would a colorblind person. Sure, from my perspective it would be displaying some superfluous information, information that I can't utilize directly. But what I could see of it would be just fine; I couldn't tell the difference.

That makes perfect sense to me, but she seems to assume she can translate her experience into a form that is viewable by trichromats. Notice the painting, "A Tetrachromat Moon." The implication is that she is showing the viewer how she sees the moon. In fact, though, that's impossible. There is no trichromat artist who, when he or she learns about color blindness, thinks they can somehow translate their experience of color into a painting a color blind person could appreciate, so it strikes me as very odd she would think she could do it for trichromats. It's too much of a no-brainer. So, she is disingenuously pretending that her condition is pretty much the same thing as synesthesia. A synesthete, could, in fact, approximate their experience for a non-synesthete in many cases. A tetrachromat, not.

So, I sum this situation up as a big mess: she doesn't seem to even understand what tetrachromacy is. That fact makes it possible for me to suspect that when she says she "sees" x number of colors in something white, she's actually referring to her artistic ability to render a solid field as composed of many colors, a device exactly like you see in the works of Monet, Renoir, Cezanne, etc. Did you notice the researchers had to propose that her "ability" was at least half dependent on training in art?

The GLIMPSE journal study (Kimberly A. Jameson, Alissa D. Winkler, Christian Herrera & Keith Goldfarb) is somewhat compelling that her tetrachromacy exists. But studies such as this, to date, are rare (is this the only one?) Perhaps time will tell when more studies are done.

One possible criticism I have of the study (unless it's my own misreading or misunderstanding), is that for its version of the minimum motion isoluminescence test, is that for different tests it uses both a neutral background and "a novel color background designed to maximally engage the fourth photoreceptor class that potential tetrachromat artist CA was presumed to phenotypically express." I would think they would have provided a spectral plot (power spectral density) of this background. But there wasn't one that I could find. I can't even find any reference in the paper what wavelength her 4th photoreceptor is hypothesized to peak at (maybe I missed it). At least they could have mentioned the peak spectral wavelength of the novel background; did I miss that too? I can't seem to find any of it.

Given her strong motivation to be "The World's First Tetrachromat Artist," and the researcher's strong motivation to be the discoverers of a human tetrachromat, I can imagine them working out a whole unconscious system of cueing each other. As they speak to her about the tests, they make it clear where a certain kind of reaction from her at a certain point, would indicate a fourth active cone, and she gets the message and complies. It would all be inexplicit and plausibly deniable. Recall the unreliability of the lie detector: when the operator thinks the suspect is guilty before the test, he unconsciously arranges the test to give results that support his opinion. So, I suspect the results of this test based on the fact both parties wanted the result they got. It has to be redone with subjects who don't know what they're being tested for from start to finish, at least. Be nice to get completely disinterested researchers as well.

It could well be she is a tetrachromat, but I am not going to buy it under these very noisy circumstances.

 

[zoobyshoe]

[Edit: Found something new. Her tests involved consecutively flickering 4 frames (two of which have time varying luminance of a particular color), in a loop, that gave the apparent illusion of rotation. So one of the rotation illusions might be better. The details of the Concetta Antico study are given here (skip to around 12:30 in the video) ps., you'll have to watch it on YouTube, the embedded version is disabled]:

O.K. so, the rotation test is to see whether a subject views a particular shade of color as closer to one of two choices, a lighter or darker choice. Is that your understanding of it?

 

[collinsmark]

I've been a little busy today. I wrote a computer program to generate a simulation of a Minimum Motion Isoluminance test like they describe in the paper. Then I captured a video of it, and made my first YouTube video! yeah!

O.K. so, the rotation test is to see whether a subject views a particular shade of color as closer to one of two choices, a lighter or darker choice. Is that your understanding of it?

Been digging my head into this, and playing around with the computer program I just wrote. The program works, and the rotation illusion is apparent.

You can think of it as 4 still frames in a continuous loop. The only thing that changes as time passes is the "variable luminance color" (see below) gradually increases in brightness the whole time.

• The first frame is a wheel having alternating spokes, one set of spokes are the "variable luminance color" and the other set are middle gray.
• The second frame is a wheel having alternating spokes, one set of spokes light gray and the other set dark gray. It's important here that the spokes in this frame are slightly rotated such that center of a given color spoke in this frame is exactly on the transition of spokes in the previous frame.
• The third frame is the same as the first, but with the "variable luminance color" and the middle gray color swapped.
• The fourth frame is the same as the second except the light and dark grays swapped.

After that everything repeats, except the "variable luminance color" gets slowly brighter as time continues.

When the "variable luminance color" is still rather dark, it gives the impression that the color is rotating in one direction (which it really isn't; it's just an illusion). But as soon as one's eyes/brain sees the luminance of the "variable luminance color" equal in luminance as the middle gray color, the perceived rotation reverses direction.

In my program I did a bit of pre-combining, and then looped each of the 4 combined frames, such that the flicker is less likely to cause seizures. Even with the combining it's still a bit weird to watch, so I added some soothing music to calm the nerves. But anyway, it works pretty well.

I'm pretty sure this was the sort of thing that was done in the Concetta Antico study. Although I dearly hope they put more effort into optimizing the color wavelengths, color saturation, and whatnot. I just threw this computer thing together in a couple of hours. It works, but I wouldn't think of using it as an actual test.



[Edit: by the way, let me explain how this relates to the study. If it was a real test, there would be a button or keypress or something that the test subject would press at the very moment she perceived the change in rotational direction. At that moment, the luminance of the "variable luminance color" is recorded.

So if you look in the test results in the paper, you'll see a graph that has luminance on the y-axis. The other graph is a comparison graph, so it has Δluminance as its y-axis. Those luminance numbers originate from the recorded points when she saw the perceived change in direction.

The bottom line that the paper is using for its conclusions is that CA seemed to perceive the rotational direction change at much less luminance (i.e., earlier, when the "variable luminance color" was dimmer) when the "variable luminance color" was in the yellow/orange/red part of the spectrum, when compared to the other test subjects.]

 

[zoobyshoe]

O.K. I've watched it about 40 times, and there's a clear and unmistakable change in the apparent direction of rotation. By stopping the video at that point, it is apparent that the threshold is when the darker green most closely matches the darker gray in value.

I'm not clear about what is being varied. At the start of the video the greens seem very much grayer than they are later, when they become more obviously "green". So, what I'm saying, I guess, is that I'm unclear about the meaning of "luminance" and how it might differ from the concept of "value" as used by artists.

It seems that for CA, the shift in rotation threshold comes when the green is darker than for other people? In other words, if she stopped the video when she saw the shift, I, or anyone else, would judge the dark green as too dark in value to be a proper match for the dark gray?

Excellent work on the "simulation"! Regardless of what I still don't understand it got me into the right ballpark.

 

[zoobyshoe]

Here are three frames from your video. The leftmost if from early in the video, and the rightmost is from late in the video. Comparing the two shows how the green changes in value over the course of the demonstration.

The middle frame is the point where I saw the rotation change direction. You can see the dark green is very close in value to the dark gray.

 

[zoobyshoe]

This is the same image as above, but changed to black and white. With no color, as a value-only comparison, you can still see all the relative values there are in the color version. So, I suspect the rotation reversal would be just as apparent, and would come at the same point, if the whole video were converted to black and white. Is there an easy way for you to do that to a video? I'd like to check that notion.

 

[collinsmark]

O.K. I've watched it about 40 times, and there's a clear and unmistakable change in the apparent direction of rotation. By stopping the video at that point, it is apparent that the threshold is when the darker green most closely matches the darker gray in value.

Yes, in my particular implementation, the reversal happens when the "test subject" (could be you, or whoever watches the video) perceives the dark green being equal luminance as the dark gray. That also the same in this case as the light green being equal intensity to the light gray. Those colors mentioned apply only to my implementation where I did the pre-combining.

Let me start by explaining the theory a little more. This explanation starts without the pre-combining. I'll come back to that later. Also, I used 16 color spokes in my video. For the sake of explanation, I'll reduce that to 4 color spokes here.

There are 4 frames repeated in succession, at a rate of roughly 5 frames per second. I'll label the frames as A, B, C, and D.

The only difference between A and C are the color positions are swapped. Similarly, the only difference between B and D are the color positions are swapped.

As successive loops continue, the only thing changing is the luminance of the "variable color" stripes in frames A and C; in this case that color is green. The mid-gray stripes in A and C do not change. Nothing changes in frames B and D. So the only things that change are the green stripes and that's it.

Note that there is no preferred motion in this sequence (clockwise or counterclockwise). It could be either way. If one perceives any rotation, it's all in the brain.

Why green? That just represents the color wavelength under test. In the CA study, they used 20 different color choices; one color for each test. In the results you'll see 20 different colors on the x-axis, each corresponding to a particular, dominant wavelength.

Of those 20 colors used in the test, 4 of them were marked with an asterisk. That's because they represented colors with a magenta hue/tint (or whatever it's called). Magenta does not exist on the rainbow spectrum. These colors marked with the asterisk were produced by combining red and blue. Why use these colors in the first place? I don't know. The authors of the study wanted to do a complete isoluminance circle around the color space, and these magenta colors are within that space, even though they cannot be pegged on the rainbow spectrum. So they included them anyway. 'Judgement call I suppose.

So anyway, I picked green for my test, just as an example test color.

Using this direct method (no combining --- I'll get to combining in a moment), the reversal happens when the subject perceives the variable color -- green here -- as being the same luminance as the mid-gray luminance.

----

Now, moving on. Those individual frames A, B, C, and D have a lot of contrast difference from one to the next. Flickering them at 5 frames per second could be quite annoying to the viewer, so what I did is pre-combined them. Instead of using the frames directly, I combed the first two frames into the first new frame, the second two frames into the second new frame, and so on. Below are the details:

E = (A+B)/2
F = (B+C)/2
G = (C+D)/2
H = (D+A)/2

[Edit: the same image as above is repeated again here for easier comparison.]

[Edit: the same image as above is repeated again here for easier comparison.]

[The image below is the sequence after combining. (i.e, the sequence is ... E ==> F==> G ==> H ==> E** ==> F** ..., where the "**" means slightly brighter luminance of the variable color)]

Ignoring nonlinearities, the combination approach doesn't change the experiment at all. It works just the same.

If I was making this for a real test, I wouldn't have combined them, or at least not so simply. The way I did it, the combinations are subject to the nonlinearities of video gamma and other nonlinear adjustments/characteristics of the video playback. [Edit: and there's also nonlinearities in biological perception too.]

But I wasn't making this for a real test. It's just a simulation, and the combination approach is a little less annoying to watch. So I went with it.

I'm not clear about what is being varied. At the start of the video the greens seem very much grayer than they are later, when they become more obviously "green". So, what I'm saying, I guess, is that I'm unclear about the meaning of "luminance" and how it might differ from the concept of "value" as used by artists.

The only thing that is being changed is the brightness of the green color. Nothing else changed. The color of the mid-gray, light-gray and dark gray (before combination) are constant colors.

After combination, you'll see two types of green: (green + light gray)/2 and (green + dark gray)/2. Only the green changes as time goes on.

Again though, my choice of "green" as the test color was pretty arbitrary, and just for demonstration purposes. If this was testing for tetrachomacy, the really interesting color to choose is a single wavelength peak around the orange part of the spectrum. Of course that cannot be generated on a typical computer monitor. The CA study repeated the testing using 20 colors overall.

It seems that for CA, the shift in rotation threshold comes when the green is darker than for other people? In other words, if she stopped the video when she saw the shift, I, or anyone else, would judge the dark green as too dark in value to be a proper match for the dark gray?

Swap "green" for "reddish and "orange" and yes. She perceived the rotation change when the variable color was dimmer, in comparison for the other test subjects. The other test subjects, when tested in that yellow-orange-red color range, didn't see the direction reversal until the variable color got much brighter, relative to CA.

That's evidence that indicates that CA has more sensitive photoreceptors for colors in that yellow-orange-red color range. At least that's the crux of the authors' arguments.

Excellent work on the "simulation"! Regardless of what I still don't understand it got me into the right ballpark.

Thanks!

 

[collinsmark]

This is the same image as above, but changed to black and white. With no color, as a value-only comparison, you can still see all the relative values there are in the color version. So, I suspect the rotation reversal would be just as apparent, and would come at the same point, if the whole video were converted to black and white. Is there an easy way for you to do that to a video? I'd like to check that notion. View attachment 84575

Yes, One could make a test where the "green" is substituted with gray (before combining). But that wouldn't isolate any particular wavelength at all. In that case, anybody and everybody would perceive the rotation change happen at the same time, regardless of their own vision or monitor settings.

In the case after combining (suppose we used the combined approach), instead of time-varying dark and light green there would be time varying dark and light gray. As soon as the time-varying dark grey equals the fixed dark gray (which would happen at the same time the time varying light-gray equals the fixed light gray) then the rotation perception would reverse.

But since it would be independent of color vision, it's not an altogether too interesting test.

[Edit: What's more than that, to show why this is the case, consider wheels A and C if we chose gray as our variable test color. Frames A and C would simply be uniformly gray circles at the perceived time of reversal, when the variable test color is the same luminance as mid-gray. At that moment, frames A and C wouldn't even contain any spokes. They would be mid-gray, all around.]

 

[zoobyshoe]

Your explanation of your video and the combined frames to avoid flickering is completely clear. Thanks.

When I express confusion about what is being varied, what I mean is what do you understand "luminance" to mean? You paraphrase it as "brightness," and didn't seem to object when I suggested you might mean what artists call "value."

The value of a color is its lightness or darkness; a completely different consideration from hue. Consider a green wall with a tree next to it. Sunlight comes through the tree onto the green wall, blocked in places by leaves but falling directly on the wall in patches where it isn't blocked. The objective color of the wall remains the same, but the shadows have a darker value of green than the directly lit patches. If you're doing a painting of that, you mix black into the green for the shadows, or mix white into it for the directly lit patches, or do both.

The reason I'm trying to nail this down is because it's apparent that the reversal of rotation is an exclusive function of value, and not hue. This is why the totally grayscale video would still demonstrate a reversal of rotation. This is of interest because the hue in question is only of importance in that it has some perceived value to the viewer. It is the value aspect of the hue, and not the hue aspect, that causes the rotation reversal.

CA seems to perceive the value of longer wavelengths to be higher than other test subjects did. In other words, a patch of magenta paint on a canvass would look to her as if it had more light falling on it than it would to one of the other subjects. Or, it would look either 1.) as if it had less black mixed into it, or 2.) as if it had more white mixed into it, (which are two ways of changing the value of a pigment). That is: she would match it in value to a gray that others would consider too light to be a value match. Which you already agreed would be the case, but I emphasize it to stay focused on the fact that this test is limited to demonstrating that she seems to have a different value perception of the longer wavelengths, not necessarily a different hue perception.

 

[collinsmark]

Your explanation of your video and the combined frames to avoid flickering is completely clear. Thanks.

When I express confusion about what is being varied, what I mean is what do you understand "luminance" to mean? You paraphrase it as "brightness," and didn't seem to object when I suggested you might mean what artists call "value."

The value of a color is its lightness or darkness; a completely different consideration from hue. Consider a green wall with a tree next to it. Sunlight comes through the tree onto the green wall, blocked in places by leaves but falling directly on the wall in patches where it isn't blocked. The objective color of the wall remains the same, but the shadows have a darker value of green than the directly lit patches. If you're doing a painting of that, you mix black into the green for the shadows, or mix white into it for the directly lit patches, or do both.

Thanks for the description of "value." I'm not so well versed in the art terms.

Yes, "brightness" is the subjective description of what I meant. I think that is subjectively kind of similar to what artists call "value."

Luminance and brightness subjectively mean the same thing. However luminance is the objective counterpart and can be measured precisely with instrumentation. For example, luminance can be measured having units of candela per square meter (cd/m²). On the other hand brightness is not so objective; brightness might be measured in percent, such as so-and-so percent of maximum brightness.

"Value," from what I gather, subjectively means the same thing, but has less objectivity. (Not to mention it applies more to pigments, and the amount of light power reflected off a pigment is mostly a function of the ambient lighting).

The term "intensity" is just as objective as "luminance," but they mean slightly different things. "Intensity" is the area under the curve of the power spectral density (PSD), where as "luminance" involves a bit of weighting before integration. Not that that really matters too much for this discussion.

The experiment under discussion involves relative values, so the distinction of these terms is not of critical importance here (although the entire experiment could be repeated with altogether lower or high luminance of all colors, to test for differences in high or low light situations). The perceived change in direction happens when the test subject perceives the luminance of the "variable luminance color" equal to the mid-gray luminance (in frames both A and C). So the "variable luminance color" luminance is relative to the mid-gray luminance, as far as this experiment is concerned. Also, CA's perception is analyzed relative to other control test subjects; so that's relative too.

By the way, let me apologize for something in my last post: I got rather sloppy with my terms. I used terms such as "variable color" when I should have said "variable luminance color." The color itself does not change (e.g., green in my simulation) only the brightness/luminance of the "variable luminance color" changes, for a given test color.

So in the GLIMPSE paper, there were a total of 20 test colors. Of those, 16 of them have a single, dominant wavelength. Within a given trial for a given test color, that wavelength is not varied; only the luminance is varied. I'm imagining light from an incandescent bulb passing through a color filter: the filter does not change, but the bulb's intensity does, as the particular trial proceeds.

The other 4 colors involve two dominant wavelengths, in order to produce the magenta colors. I have no idea which particular wavelengths they are (if they're specified in the paper I cannot find them), but suffice it to say those wavelengths are on the red and blue part of the spectrum, and do not change for a given test color.

So the algorithm for the entire experiment, from what I gather, might be something like this:

1. Pick one of the 20 test colors and configure apparatus appropriately.
2. Put test subject in the test apparatus (whatever that may be.. possibly involving restricted head movement, controlled setting, etc).
   
3. Gradually increase luminance of the test color (i.e., "variable luminance color") while keeping all background colors constant (constant luminance colors such as the mid-gray in frames A and C, and anything in frames B and D).
   
4. When the test subject presses the button indicating direction reversal, record the luminance value of the test color ("variable luminance color").
5. Repeat steps 1-4 for all 20 test colors.
6. (Optional) Repeat steps 1-5 with different luminance of the constant gray colors, to test for differences in high- and low-light environments.
7. (Optional) Repeat steps 1-6, but substitute mid-gray background with "novel background color."
   
8. Repeat steps 1-7 for the purposes of averaging and repeatability statistics.

[Edit: According to Section 3 of the Article Supplement, something like steps 6 and 7 were in fact performed. The total number judgments per test subject was stated to be "~7000 isoluminance judgments (plus practice and initiation trials)."]

The reason I'm trying to nail this down is because it's apparent that the reversal of rotation is an exclusive function of value, and not hue. This is why the totally grayscale video would still demonstrate a reversal of rotation. This is of interest because the hue in question is only of importance in that it has some perceived value to the viewer. It is the value aspect of the hue, and not the hue aspect, that causes the rotation reversal.

Yes, that's right. "Value" here being a subjective term meaning relative brightness or relative luminance.

CA seems to perceive the value of longer wavelengths to be higher than other test subjects did. In other words, a patch of magenta paint on a canvass would look to her as if it had more light falling on it than it would to one of the other subjects. Or, it would look either 1.) as if it had less black mixed into it, or 2.) as if it had more white mixed into it, (which are two ways of changing the value of a pigment). That is: she would match it in value to a gray that others would consider too light to be a value match. Which you already agreed would be the case, but I emphasize it to stay focused on the fact that this test is limited to demonstrating that she seems to have a different value perception of the longer wavelengths, not necessarily a different hue perception.

I should comment here about an important difference between paint and light. Increasing the luminance of a test color, for the purposes of this experiment, does not mean adding white light. Superimposing white light onto the test color would increase the bandwidth of the overall color, and there is no evidence of that in the experimental description. Rather, think of it as a colored light bulb (or a standard light bulb with a colored filter on the exterior) that has variable intensity.

Other than that, yes, I agree with you completely. For any particular test color (of the 20 used), the test color's hue does not change. Only the color's intensity/luminance changes. If you were to take the power spectral density of the test color, its shape would remain constant for a particular test color: only its magnitude changes as the test proceeds.

So why did CA also excel in her perception of the magenta colors (not just the orange part of the spectrum)? We can't say without knowing more about the particular dominant wavelengths and their respective bandwidths in the four magenta test colors. But presumably, if those test colors had some spectral energy in the red/orange part of the spectrum (in addition to the second, blue wavelength) then that might give CA an advantage if her 4th color receptor was excited (due to the spectral energy in the corresponding red/orange part). But since the details of the two dominate wavelengths were not given (as far as I can find anyway), I can't really comment on those.

At least that's my take on it, as far as I can figure out.

---

Which brings me to an idea about substituting that "novel background" they mentioned for mid-gray. If the mid-gray color in frames A and C was substituted for an orange color, having 2 dominate wavelengths in the red and green part of the spectrum, although the test color ("variable wavelength color") had a single wavelength in that orange part of the spectrum (metamers), that would make an excellent test for tetrachromacy, me thinks.

It would be like the grayscale version we talked about, except for a difference that only tetrachromats could see.

[Edit: although this difference, as recorded in this experiment, would not necessarily distinguish between strong and weak tetrachromats. The difference in isoluminance perception might register just as well between weak tetrachromats and tricrhomats as it does between strong tetrachromats and tricrhomats. (where a "strong" tetrachromat perceives a distinctly unique color, whereas a "weak" tetrachromat registers the different isoluminance perception, but does not perceive a wholly unique color.)]

Is that how they used the "novel background"? I don't know. I can't seem to find such nitty-gritty details in the paper. (The information that I cannot seem to find is the spectral properties of this novel background color.)

 

[dragoneyes001]

got a measly 41 but two things applied to the poor score. I'm on a touchscreen laptop so the reflection from the glass makes any straining to see something hard on the eyes (too much glare) as well as my not trying to get a perfect score (only wanted to try the test not spend the time to ace it. so let things pass as good enough)

 

[jesssmart]

Thanx! Interesting test. I got 7!