PS CS3 puzzler – non-linear behaviour in gradients

Alan Browne wrote:

Can anyone shed light on this?

1) In PS, follow the instructions at:

http://www.luminous-landscape.com/essays/test-charts.shtml and generate the Granger chart as shown. Oddly, it has a strange distribution given that the gradient layer is linear top-to-bottom. What is creating those diagonals in the chart?

2) For comparison, see;

http://farm3.static.flickr.com/2790/4337946696_8ef5e104ff_o. jpg

Perhaps simply because the individual gradients themselves are non-linear, compared to gradients applied by other photo-editing programs. Gimp "colorcube analysis" shows the non-linearity of the gradients here: http://i49.tinypic.com/25jxge9.png
Gradients produced in Gimp (the same as gradients produced in other programs) are on the left).

On Mon, 08 Feb 2010 09:34:07 +1300, Me wrote:

Alan Browne wrote:

Can anyone shed light on this?

1) In PS, follow the instructions at:

http://www.luminous-landscape.com/essays/test-charts.shtml and generate the Granger chart as shown. Oddly, it has a strange distribution given that the gradient layer is linear top-to-bottom. What is creating those diagonals in the chart?

2) For comparison, see;

http://farm3.static.flickr.com/2790/4337946696_8ef5e104ff_o. jpg

Perhaps simply because the individual gradients themselves are non-linear, compared to gradients applied by other photo-editing programs. Gimp "colorcube analysis" shows the non-linearity of the gradients here: http://i49.tinypic.com/25jxge9.png
Gradients produced in Gimp (the same as gradients produced in other programs) are on the left).

Photoshop can provide linear gradients, but not by default. Use Photoshop’s gradient editor, set the smoothing parameter to zero, and no dither. Why did they do it this way? Why ask why.
—
Mike Russell – http://www.curvemeister.com

Mike Russell wrote:

On Mon, 08 Feb 2010 09:34:07 +1300, Me wrote:

Alan Browne wrote:

Can anyone shed light on this?

1) In PS, follow the instructions at:

http://www.luminous-landscape.com/essays/test-charts.shtml and generate the Granger chart as shown. Oddly, it has a strange distribution given that the gradient layer is linear top-to-bottom. What is creating those diagonals in the chart?

2) For comparison, see;

http://farm3.static.flickr.com/2790/4337946696_8ef5e104ff_o. jpg

Perhaps simply because the individual gradients themselves are non-linear, compared to gradients applied by other photo-editing programs. Gimp "colorcube analysis" shows the non-linearity of the gradients here: http://i49.tinypic.com/25jxge9.png
Gradients produced in Gimp (the same as gradients produced in other programs) are on the left).

Photoshop can provide linear gradients, but not by default. Use Photoshop’s gradient editor, set the smoothing parameter to zero, and no dither. Why did they do it this way? Why ask why.

Thanks – that works – problem solved.
Now perhaps someone can advise Ludicrous Landscape’s editors with instructions on how to make a "proper" Granger chart.

On 10-02-07 19:40 , Mike Russell wrote:

On Mon, 08 Feb 2010 09:34:07 +1300, Me wrote:

Alan Browne wrote:

Can anyone shed light on this?

1) In PS, follow the instructions at:

http://www.luminous-landscape.com/essays/test-charts.shtml and generate the Granger chart as shown. Oddly, it has a strange distribution given that the gradient layer is linear top-to-bottom. What is creating those diagonals in the chart?

2) For comparison, see;

http://farm3.static.flickr.com/2790/4337946696_8ef5e104ff_o. jpg

Perhaps simply because the individual gradients themselves are non-linear, compared to gradients applied by other photo-editing programs. Gimp "colorcube analysis" shows the non-linearity of the gradients here: http://i49.tinypic.com/25jxge9.png
Gradients produced in Gimp (the same as gradients produced in other programs) are on the left).

Photoshop can provide linear gradients, but not by default. Use Photoshop’s gradient editor, set the smoothing parameter to zero, and no dither. Why did they do it this way? Why ask why.

Doesn’t work here – I turned off dither (for spectrum and gradient) and I set the smooth to zero for both and for either one. I get the same non-linear plot.

Fresh eyes tomorrow.

—
gmail originated posts are filtered due to spam.

On 10-02-07 19:40 , Mike Russell wrote:

Photoshop can provide linear gradients, but not by default. Use Photoshop’s gradient editor, set the smoothing parameter to zero, and no dither. Why did they do it this way? Why ask why.

Hi Mike, I followed your instructions, but I do not get the ‘standard’ Granger. I set no dither for both the spectrum and the gradient. I set smoothing to 0 (for either and both). But I do not get the Granger w/o the artifacts. Any idea what I’m missing?

(If I do this with "HardLight" instead of Luminosity, it does work).

—
gmail originated posts are filtered due to spam.

On Tue, 09 Feb 2010 16:14:44 -0500, Alan Browne wrote:

On 10-02-07 19:40 , Mike Russell wrote:

Photoshop can provide linear gradients, but not by default. Use Photoshop’s gradient editor, set the smoothing parameter to zero, and no dither. Why did they do it this way? Why ask why.

Hi Mike, I followed your instructions, but I do not get the ‘standard’ Granger. I set no dither for both the spectrum and the gradient. I set smoothing to 0 (for either and both). But I do not get the Granger w/o the artifacts. Any idea what I’m missing?

(If I do this with "HardLight" instead of Luminosity, it does work).

Photoshop is giving an accurate result in both cases. The hard light layer will darken RGB values of the target layer for values less than 128, and lighten it for values greater than 128. The result is an artifact free Granger chart, similar to what you see from other products.

Luminosity is a hue dependent function, weighted by the luminosity values, with b the darkest and green the lightest of the rgb primaries. It’s not possible, in RGB space, for a pure blue to have a luminosity greater than 11%. Above that it is necessary to add white to the blue to get the value dictated by the luminosity layer.

The zig zag artifact has six peaks, one for each primary of the HSL hexcone. Another way to think of it is a plot of the luminosity of various mixtures of r, g, and b, with blue being the darkest, and yellow being the lightest. Or think of it as a flattened cylindrical projection of the HSL hexcone.

Jacob Rus has recently done a wiki page on HSL and HSB. The "set height from luma" operation shown in the geometric derivation illustration is what is happening with Photoshop’s luminosity mode.
—
Mike Russell – http://www.curvemeister.com

On 10-02-09 23:52 , Mike Russell wrote:

On Tue, 09 Feb 2010 16:14:44 -0500, Alan Browne wrote:

On 10-02-07 19:40 , Mike Russell wrote:

Photoshop can provide linear gradients, but not by default. Use Photoshop’s gradient editor, set the smoothing parameter to zero, and no dither. Why did they do it this way? Why ask why.

Hi Mike, I followed your instructions, but I do not get the ‘standard’ Granger. I set no dither for both the spectrum and the gradient. I set smoothing to 0 (for either and both). But I do not get the Granger w/o the artifacts. Any idea what I’m missing?

(If I do this with "HardLight" instead of Luminosity, it does work).

Photoshop is giving an accurate result in both cases. The hard light layer will darken RGB values of the target layer for values less than 128, and lighten it for values greater than 128. The result is an artifact free Granger chart, similar to what you see from other products.
Luminosity is a hue dependent function, weighted by the luminosity values, with b the darkest and green the lightest of the rgb primaries. It’s not possible, in RGB space, for a pure blue to have a luminosity greater than 11%. Above that it is necessary to add white to the blue to get the value dictated by the luminosity layer.

The zig zag artifact has six peaks, one for each primary of the HSL hexcone. Another way to think of it is a plot of the luminosity of various mixtures of r, g, and b, with blue being the darkest, and yellow being the lightest. Or think of it as a flattened cylindrical projection of the HSL hexcone.

I sort of got all that from the result (without the fine disection you give it of course).

Jacob Rus has recently done a wiki page on HSL and HSB. The "set height from luma" operation shown in the geometric derivation illustration is what is happening with Photoshop’s luminosity mode.

I really wanted to know why if I’m following your instructions, I’m not getting your result. Someone else did. ("Me" but not me).

eg: I’m doing something wrong in the process.

—
gmail originated posts are filtered due to spam.

On Tue, 09 Feb 2010 23:56:35 -0500, Alan Browne wrote:

I really wanted to know why if I’m following your instructions, I’m not getting your result. Someone else did. ("Me" but not me).
eg: I’m doing something wrong in the process.

I think they were referring to the gradient not having a clean histogram, rather than the result with the granger chart. The granger chart will have the artifacts you mention for Luminosity, because of the relationship of hue and luminosity, but not for other modes such as hard light. —
Mike Russell – http://www.curvemeister.com

On 10-02-10 17:25 , Mike Russell wrote:

On Tue, 09 Feb 2010 23:56:35 -0500, Alan Browne wrote:

I really wanted to know why if I’m following your instructions, I’m not getting your result. Someone else did. ("Me" but not me).
eg: I’m doing something wrong in the process.

I think they were referring to the gradient not having a clean histogram, rather than the result with the granger chart. The granger chart will have the artifacts you mention for Luminosity, because of the relationship of hue and luminosity, but not for other modes such as hard light.

You’re not hearing me.

"Me" (a different poster) followed your suggestion for turning off dithering and setting smoothing to 0. Using Luminosity He got a nominally correct Granger.

I (that is me, not "Me") followed your instructions and did not get the nominally correct Granger using Luminosity (I did with "Hardlight").

What could account for the different result?

—
gmail originated posts are filtered due to spam.

On Wed, 10 Feb 2010 18:15:55 -0500, Alan Browne wrote:

On 10-02-10 17:25 , Mike Russell wrote:

On Tue, 09 Feb 2010 23:56:35 -0500, Alan Browne wrote:

I really wanted to know why if I’m following your instructions, I’m not getting your result. Someone else did. ("Me" but not me).
eg: I’m doing something wrong in the process.

I think they were referring to the gradient not having a clean histogram, rather than the result with the granger chart. The granger chart will have the artifacts you mention for Luminosity, because of the relationship of hue and luminosity, but not for other modes such as hard light.

You’re not hearing me.

"Me" (a different poster) followed your suggestion for turning off dithering and setting smoothing to 0. Using Luminosity He got a nominally correct Granger.

I (that is me, not "Me") followed your instructions and did not get the nominally correct Granger using Luminosity (I did with "Hardlight").
What could account for the different result?

I think I’m hearing both you, and Me. I guess we’ll just have to wait for Me to clear up the matter.
—
Mike Russell – http://www.curvemeister.com

On 2/9/2010 8:52 PM, Mike Russell wrote:

On Tue, 09 Feb 2010 16:14:44 -0500, Alan Browne wrote:

On 10-02-07 19:40 , Mike Russell wrote:

Photoshop can provide linear gradients, but not by default. Use Photoshop’s gradient editor, set the smoothing parameter to zero, and no dither. Why did they do it this way? Why ask why.

Hi Mike, I followed your instructions, but I do not get the ‘standard’ Granger. I set no dither for both the spectrum and the gradient. I set smoothing to 0 (for either and both). But I do not get the Granger w/o the artifacts. Any idea what I’m missing?

(If I do this with "HardLight" instead of Luminosity, it does work).

Photoshop is giving an accurate result in both cases. The hard light layer will darken RGB values of the target layer for values less than 128, and lighten it for values greater than 128. The result is an artifact free Granger chart, similar to what you see from other products.

Hard light is not exactly the same, I managed to make a matching chart with Screen mode above 128 and Multiply mode below 128. That’s what most programs call luminosity. The question is why the different approaches?

Luminosity is a hue dependent function, weighted by the luminosity values, with b the darkest and green the lightest of the rgb primaries. It’s not possible, in RGB space, for a pure blue to have a luminosity greater than 11%. Above that it is necessary to add white to the blue to get the value dictated by the luminosity layer.

The zig zag artifact has six peaks, one for each primary of the HSL hexcone. Another way to think of it is a plot of the luminosity of various mixtures of r, g, and b, with blue being the darkest, and yellow being the lightest. Or think of it as a flattened cylindrical projection of the HSL hexcone.

Jacob Rus has recently done a wiki page on HSL and HSB. The "set height from luma" operation shown in the geometric derivation illustration is what is happening with Photoshop’s luminosity mode.

http://en.wikipedia.org/wiki/HSL_and_HSV
That first illustration looks a lot like what we are discussing here.

Math for luma figured into the best explanation I could find. That explained the asymetrical humps in the PS version with ‘luminance’ but what is the scheme that shows symmetrical peaks in the diagram and why did the other programs chose that math for what they call luminance? The big difference I saw was that if you convert to grayscale, the PS version goes to a smooth transition where the other shows the colors for their lightness: yellow is light, blue is dark, green and red intermediate. The PS version preserves grayscale lightness/darkness as can be seen in the last frame of the set below.

Here’s examples and the blending modes used to achieve them: http://www.flickr.com/photos/edgehill/sets/72157623375625518 / I also tried it in lab mode, out of curiosity.
#2 was pasted together from using those two blending modes for top & bottom, then squishing them back to fit in a square. Most programs work this way for what they call luminance. #1 is the default PS behavior.

At first I was just curious but I do use the luminance blending mode pretty often on contrast type adjustment layers so that I can tweak curves or levels without changing saturation. Normally a contrast increase will give a more saturated image and I often like to keep things subtle. I’m aware that lab mode is the better way to do this but it’s a hassle and I’m just not in the habit of working in lab. Is the luminosity blend mode going to mess up my colors? Like if curves darken a blue area, it looks like that might turn pale. Or maybe this really is the better way to do it and actually preserves luminosity better.

On Thu, 11 Feb 2010 11:02:52 -0800, Paul Furman wrote:

On 2/9/2010 8:52 PM, Mike Russell wrote:

On Tue, 09 Feb 2010 16:14:44 -0500, Alan Browne wrote:

On 10-02-07 19:40 , Mike Russell wrote:

Photoshop can provide linear gradients, but not by default. Use Photoshop’s gradient editor, set the smoothing parameter to zero, and no dither. Why did they do it this way? Why ask why.

Hi Mike, I followed your instructions, but I do not get the ‘standard’ Granger. I set no dither for both the spectrum and the gradient. I set smoothing to 0 (for either and both). But I do not get the Granger w/o the artifacts. Any idea what I’m missing?

(If I do this with "HardLight" instead of Luminosity, it does work).

Photoshop is giving an accurate result in both cases. The hard light layer will darken RGB values of the target layer for values less than 128, and lighten it for values greater than 128. The result is an artifact free Granger chart, similar to what you see from other products.

Hard light is not exactly the same, I managed to make a matching chart with Screen mode above 128 and Multiply mode below 128. That’s what most programs call luminosity. The question is why the different approaches?

Luminosity is a hue dependent function, weighted by the luminosity values, with b the darkest and green the lightest of the rgb primaries. It’s not possible, in RGB space, for a pure blue to have a luminosity greater than 11%. Above that it is necessary to add white to the blue to get the value dictated by the luminosity layer.

The zig zag artifact has six peaks, one for each primary of the HSL hexcone. Another way to think of it is a plot of the luminosity of various mixtures of r, g, and b, with blue being the darkest, and yellow being the lightest. Or think of it as a flattened cylindrical projection of the HSL hexcone.

Jacob Rus has recently done a wiki page on HSL and HSB. The "set height from luma" operation shown in the geometric derivation illustration is what is happening with Photoshop’s luminosity mode.

http://en.wikipedia.org/wiki/HSL_and_HSV
That first illustration looks a lot like what we are discussing here.

Math for luma figured into the best explanation I could find. That explained the asymetrical humps in the PS version with ‘luminance’ but what is the scheme that shows symmetrical peaks in the diagram and why did the other programs chose that math for what they call luminance?

I think the other programs simply multiplied the RGB channels equally by the grayscale value over 255.

The
big difference I saw was that if you convert to grayscale, the PS version goes to a smooth transition where the other shows the colors for their lightness: yellow is light, blue is dark, green and red intermediate. The PS version preserves grayscale lightness/darkness as can be seen in the last frame of the set below.

Here’s examples and the blending modes used to achieve them: http://www.flickr.com/photos/edgehill/sets/72157623375625518 / I also tried it in lab mode, out of curiosity.
#2 was pasted together from using those two blending modes for top & bottom, then squishing them back to fit in a square. Most programs work this way for what they call luminance. #1 is the default PS behavior.
At first I was just curious but I do use the luminance blending mode pretty often on contrast type adjustment layers so that I can tweak curves or levels without changing saturation. Normally a contrast increase will give a more saturated image and I often like to keep things subtle. I’m aware that lab mode is the better way to do this but it’s a hassle and I’m just not in the habit of working in lab. Is the luminosity blend mode going to mess up my colors? Like if curves darken a blue area, it looks like that might turn pale. Or maybe this really is the better way to do it and actually preserves luminosity better.

It all depends on the image – the important thing is to be quick enough in your workflow to try it one way, then another, and compare results. Dan Margulis’s Picture Postcard workflow is an example of this, though it is done with a combination of curves and Photoshop’s apply image command.

Speaking of just blues in the context of sky – most viewers like skies to be dark and saturated, so yes, applying a luminosity layer to a blue sky would generally lighten it. In this case you would use apply image, uaing the red channel in luminosity mode, and use a curve to boost the effect further. If there are important foreground elements, then use the green or blue channels (or the b channel in Lab) as a mask.

There are dozens of ways to use curves, masking, apply image, layer blending, and other functions, each of which adds a bit of punch to your image as far as color or detail.

BTW – on your flickr page you pose the question of why use a Granger chart at all – for me it’s a curiousity item that is helpful for understanding the RGB color space, but nothing more than that. As an artificial gradient, it is of no relevance to photography or printing related to photography.
—
Mike Russell – http://www.curvemeister.com

On 2/11/2010 10:00 PM, Mike Russell wrote:

On Thu, 11 Feb 2010 11:02:52 -0800, Paul Furman wrote:

On 2/9/2010 8:52 PM, Mike Russell wrote:

On Tue, 09 Feb 2010 16:14:44 -0500, Alan Browne wrote:

On 10-02-07 19:40 , Mike Russell wrote:

Photoshop can provide linear gradients, but not by default. Use Photoshop’s gradient editor, set the smoothing parameter to zero, and no dither. Why did they do it this way? Why ask why.

Hi Mike, I followed your instructions, but I do not get the ‘standard’ Granger. I set no dither for both the spectrum and the gradient. I set smoothing to 0 (for either and both). But I do not get the Granger w/o the artifacts. Any idea what I’m missing?

(If I do this with "HardLight" instead of Luminosity, it does work).

Photoshop is giving an accurate result in both cases. The hard light layer will darken RGB values of the target layer for values less than 128, and lighten it for values greater than 128. The result is an artifact free Granger chart, similar to what you see from other products.

Hard light is not exactly the same, I managed to make a matching chart with Screen mode above 128 and Multiply mode below 128. That’s what most programs call luminosity. The question is why the different approaches?

Luminosity is a hue dependent function, weighted by the luminosity values, with b the darkest and green the lightest of the rgb primaries. It’s not possible, in RGB space, for a pure blue to have a luminosity greater than 11%. Above that it is necessary to add white to the blue to get the value dictated by the luminosity layer.

The zig zag artifact has six peaks, one for each primary of the HSL hexcone. Another way to think of it is a plot of the luminosity of various mixtures of r, g, and b, with blue being the darkest, and yellow being the lightest. Or think of it as a flattened cylindrical projection of the HSL hexcone.

Jacob Rus has recently done a wiki page on HSL and HSB. The "set height from luma" operation shown in the geometric derivation illustration is what is happening with Photoshop’s luminosity mode.

http://en.wikipedia.org/wiki/HSL_and_HSV
That first illustration looks a lot like what we are discussing here.

Math for luma figured into the best explanation I could find. That explained the asymetrical humps in the PS version with ‘luminance’ but what is the scheme that shows symmetrical peaks in the diagram and why did the other programs chose that math for what they call luminance?

I think the other programs simply multiplied the RGB channels equally by the grayscale value over 255.

The
big difference I saw was that if you convert to grayscale, the PS version goes to a smooth transition where the other shows the colors for their lightness: yellow is light, blue is dark, green and red intermediate. The PS version preserves grayscale lightness/darkness as can be seen in the last frame of the set below.

Here’s examples and the blending modes used to achieve them: http://www.flickr.com/photos/edgehill/sets/72157623375625518 / I also tried it in lab mode, out of curiosity.
#2 was pasted together from using those two blending modes for top& bottom, then squishing them back to fit in a square. Most programs work this way for what they call luminance. #1 is the default PS behavior.
At first I was just curious but I do use the luminance blending mode pretty often on contrast type adjustment layers so that I can tweak curves or levels without changing saturation. Normally a contrast increase will give a more saturated image and I often like to keep things subtle. I’m aware that lab mode is the better way to do this but it’s a hassle and I’m just not in the habit of working in lab. Is the luminosity blend mode going to mess up my colors? Like if curves darken a blue area, it looks like that might turn pale. Or maybe this really is the better way to do it and actually preserves luminosity better.

It all depends on the image – the important thing is to be quick enough in your workflow to try it one way, then another, and compare results. Dan Margulis’s Picture Postcard workflow is an example of this, though it is done with a combination of curves and Photoshop’s apply image command.
Speaking of just blues in the context of sky – most viewers like skies to be dark and saturated, so yes, applying a luminosity layer to a blue sky would generally lighten it. In this case you would use apply image, uaing the red channel in luminosity mode, and use a curve to boost the effect further. If there are important foreground elements, then use the green or blue channels (or the b channel in Lab) as a mask.

There are dozens of ways to use curves, masking, apply image, layer blending, and other functions, each of which adds a bit of punch to your image as far as color or detail.

BTW – on your flickr page you pose the question of why use a Granger chart at all – for me it’s a curiousity item that is helpful for understanding the RGB color space, but nothing more than that. As an artificial gradient, it is of no relevance to photography or printing related to photography.

Thanks! I’ve been discovering snapshots on the history toolbar are good for that filp-back-n-forth type comparison; an effective way to evaluate complex options.

I take it you mean, if blues are getting washed out, then just work on the red (and or green) channel of an adjustment layer (when setting luminance blend mode to avoid saturation changes as I described). I tried that on one typical sort of landscape with a washed out sky and all it did was emphasize the dark vignetting in the corners, so yeah, it all depends on the image and any number of tweaks can be applied till it looks right.

The value of the chart for general comprehension is like this question – looking at a few of these charts flipping through the blend modes has helped me understand blend modes and what to look for.

Mike Russell wrote:

On Wed, 10 Feb 2010 18:15:55 -0500, Alan Browne wrote:

On 10-02-10 17:25 , Mike Russell wrote:

On Tue, 09 Feb 2010 23:56:35 -0500, Alan Browne wrote:

I really wanted to know why if I’m following your instructions, I’m not getting your result. Someone else did. ("Me" but not me).
eg: I’m doing something wrong in the process.

I think they were referring to the gradient not having a clean histogram, rather than the result with the granger chart. The granger chart will have the artifacts you mention for Luminosity, because of the relationship of hue and luminosity, but not for other modes such as hard light.

You’re not hearing me.

"Me" (a different poster) followed your suggestion for turning off dithering and setting smoothing to 0. Using Luminosity He got a nominally correct Granger.

I (that is me, not "Me") followed your instructions and did not get the nominally correct Granger using Luminosity (I did with "Hardlight").
What could account for the different result?

I think I’m hearing both you, and Me. I guess we’ll just have to wait for Me to clear up the matter.

You solved the problem with non-linear gradients.
The relationship between luminosity and hue in photoshop also explains why the method suggested by Luminous Landscape fails.
It remains a mystery why Luminous Landscape didn’t see that their "Granger chart" was a shameful mistake, and a further mystery – to me anyway – why they thought that a Granger chart was useful, except as a curiosity. They also failed to suggest that using jpeg compression on such a chart was a waste of time and to be seriously avoided, and why on the sort of monitors many people use these days, correct display of gradients can be a big problem.
But from years of reading articles about photography on Luminous Landscape, I’d come to the conclusion that they are a pretty hapless lot of obsessive gadget-geeks, not real photographers.

Alan Browne wrote:

1) In PS, follow the instructions at:
http://www.luminous-landscape.com/essays/test-charts.shtmlan d generate the Granger chart as shown.  Oddly, it has a strange distribution given that the gradient layer is linear top-to-bottom.  What is creating those diagonals in the chart?

As Mike mentioned, this is because Photoshop’s hue/saturation/color/ luminosity blend modes are with reference to what my Wikipedia article about HSL & HSV [1] calls a "luma/chroma/hue" model. If you look at the second image in that article, it’s the shape at the bottom right. What Photoshop’s blend modes call “luminosity” is close to .3*R + .6*G + .1*B, what the blend modes call “saturation” is max(R, G, B) – min(R, G, B) — notice that this is different from both HSL "saturation" and HSV "saturation" — and what the blend modes call hue is the same as HSL/HSV hue.

[1]: http://en.wikipedia.org/wiki/HSL_and_HSV

So anyway, that "granger chart" (I’ve never heard of such a thing, and can’t imagine a use for it; I wonder how the luminous landscape guys came up with it) basically shows what would happen if you stretched out the odd asymmetrical bicone shape of the luma/chroma/hue model into a cylinder. In other words, for each combination of hue and luma, it shows the color with maximum chroma.

Perhaps the purpose of this is to aid selection of as-chromatic-as- possible colors, based on some scheme in hue/lightness? If so, this isn’t such a great tool, and really highlights the problems of using luma as a proxy for lightness (e.g. CIELAB L* or CIECAM02 J): luma is derived directly from the gamma-corrected R, G, B values stored in an image in a space like Adobe RGB or sRGB, but human brightness/ lightness perception is a non-linear adjustment to the direct luminance (Y) signal taken in by the eyes. Basically, using luma as a proxy for lightness leads to great distortions for bright colors. This is dramatically demonstrated by the “Granger chart”, which has bright diagonal bands along the highest-chroma edges. If luma were a good proxy for lightness, this chart would look pretty smooth.

It’s also problematic to use this definition of "hue" based on RGB primaries, because it’s basically arbitrary. Much better would be to space hue as in Munsell, CIELAB, or CIECAM02 geometries.

So anyway, again, I can’t think of a good use for this "Granger chart", and I recommend ignoring/avoiding it.

2) For comparison, see;

http://farm3.static.flickr.com/2790/4337946696_8ef5e104ff_o. jpg

I’m not sure what this is about. Since I’ve never heard of a "Granger chart" before today, I’m not sure what a proper one looks like, but it’s not at all clear to me that the results from these other editors (the edge of an HSL cylinder) is it. Presumably the luminous landscape people realize the difference.

If you want to make a chart like this using photoshop, follow these steps:
1) Make a horizontal gradient (smoothness = 0) where red, yellow, green, cyan, blue, magenta, red are at 0%, 16%, 33%, 50%, 67%, 83%
2) Add a layer above, and make vertical gradient from bottom to top
which has colors (bottom of the gradient interface) go black, black, white, white at 0%, 50%, 50%, 100% (the two 50%s should be as close as you can make them, like 1 pixel apart), and which has transparency (top of the gradient interface) opaque (100% opacity), transparent (0%), opaque, at positions 0%, 50%, 100%. Make this gradient still have smoothness = 0. Leave the layer in normal mode, since we just want to interpolate linearly between red, yellow, etc. and black at the bottom or white at the top.

Do those instructions make sense?

* * *

All of this is a completely separate issue from photoshop’s non-linear gradients. I generally turn smoothness on my gradients down to 0, to make them linear, because I tend to use gradients for making charts/ graphics rather than for editing photos. I’m not sure quite how they decided on the interpolation method for the 100% smoothness version. I think it’s the same as the interpolation used by the Curves tool (and I don’t understand the particular choice of math used there, either — it’s not the best for the job).

Hope that helps,
Jacob

Incidentally, in case anyone wants a "proper Granger rainbow" (which is to say, the edge of an HSL cylinder) of their own, without following the steps above, I’ve uploaded one here:
http://www.mediafire.com/?2mmam2mdyzt

Just resize this photoshop file to any desired dimensions.

Cheers,
Jacob

On Mon, 8 Mar 2010 15:42:18 -0800 (PST), Jacob Rus wrote:

Incidentally, in case anyone wants a "proper Granger rainbow" (which is to say, the edge of an HSL cylinder) of their own, without following the steps above, I’ve uploaded one here:
http://www.mediafire.com/?2mmam2mdyzt

Just resize this photoshop file to any desired dimensions.
Cheers,
Jacob

May not be useful, but it sure looks pretty. Reminds me of the rainbow I saw on the way back from work today.

Interesting that there are relatively bright vertical lines at the CMY hues, but not at the RGB ones. I’ll venture a guess that this is because these hues have two sets of pixels contributing photons, instead of only one.

So, for example RGB(255,0,0) will be less bright than RGB(255,255,0) or RGB(255,0,255).
—
Mike Russell – http://www.curvemeister.com

Mike Russell wrote:

Interesting that there are relatively bright vertical lines at the CMY hues, but not at the RGB ones.  I’ll venture a guess that this is because these hues have two sets of pixels contributing photons, instead of only one.  

So, for example RGB(255,0,0) will be less bright than RGB(255,255,0) or RGB(255,0,255).

Pretty much. If you look at the deformations that go into making the HSL cylinder from an RGB cube – in the diagram <http:// en.wikipedia.org/wiki/File:Hsl-and-hsv.svg> look at the tilted cube in the top middle and notice that the following step pushes red, green, and blue *up*, while pushing cyan, yellow, and magenta *down* – these bright vertical lines are the top three edges of the cube (stretching between white and C, M, or Y). At red, green, and blue, there are also vertical lines, but they’re darker than the surrounding area, rather than lighter.

Cheers,
Jacob

On Wed, 10 Mar 2010 02:23:48 -0800 (PST), Jacob Rus wrote:

Mike Russell wrote:

Interesting that there are relatively bright vertical lines at the CMY hues, but not at the RGB ones.  I’ll venture a guess that this is because these hues have two sets of pixels contributing photons, instead of only one.  

So, for example RGB(255,0,0) will be less bright than RGB(255,255,0) or RGB(255,0,255).

Pretty much. If you look at the deformations that go into making the HSL cylinder from an RGB cube – in the diagram <http:// en.wikipedia.org/wiki/File:Hsl-and-hsv.svg> look at the tilted cube in the top middle and notice that the following step pushes red, green, and blue *up*, while pushing cyan, yellow, and magenta *down* – these bright vertical lines are the top three edges of the cube (stretching between white and C, M, or Y). At red, green, and blue, there are also vertical lines, but they’re darker than the surrounding area, rather than lighter.

Cheers,
Jacob

Ah – hadn’t quite pieced that together. Thanks. This also explains the varying width of the colored bands for RGB and CMY on the default "rainbow" gradient, and the rainbow pattern on the color picker hue slider.

BTW – I just noticed that there is a mini-Granger chart along the bottom of the color palette. I believe Kai Krause’s color picker also used a Granger-style pattern.
—
Mike Russell – http://www.curvemeister.com