skip to main content

Exploring the LCH and HSV blend modes using a Wikipedia photograph of Kenyon Cox's portrait of Saint-Gaudens

This two-part tutorial was inspired by an email conversation with Americo, who suggested to compare using the LCH and HSV color spaces to determine the color palette of Kenyon Cox's portait of Saint-Gaudens. Part 1 (what you are reading right now) has two goals:

  1. Provide an easy way to explore and compare the LCH and HSV blend modes.
  2. Take the first step in analyzing Kenyon Cox's portrait of Saint-Gaudens, in which we learn that the Wikipedia photograph of Cox's portrait is not a very good photograph for learning anything useful about Cox's actual color palette.

Part 2 will take a look at the Met Museum's presumably more colorimetrically accurate digital version of Cox's portrait.

Written December 2016. September 2017 — this page needs some revision to account for the fact that in GIMP 2.8 and 2.9, there is no HSV color blend mode — what I assumed was HSV color blend is actually HSL color blend — the other HS"X" blend modes really are HSV blend modes.

The inspiration for this tutorial, and how to follow along

An email discussion about a portrait by Kenyon Cox

Americo and I were having an email discussion about a portrait of the American sculptor Augustus Saint-Gaudens. Here's an excerpt from our conversation:

elle: I spent some time this morning rambling over the internet looking at art, starting from Fra_Angelico. Somehow I ended up at an article on Augustus Saint-Gaudens, which had a link to a portrait of Saint-Gaudens done by Kenyon Cox:
Kenyon Cox portrait of Saint-Gaudens
There's a large original file, and looking at it, I thought the colors and shapes and edges were really quite beautiful. I was wondering what your reactions might be. americo: The portrait is a good exercise of a limited palette over ocher and probably some reddish... I don't remember all pigment names... but I think that exists red pigments with ascendance versus ocher tones. This work reminds me of some painting portraits by Anders Zorn, who also often used a very limited palette for figure and portrait paintings. Reading the Wikipedia Cox biography, I see that he was very interested in the line drawing and well finished before painting. So this approach, we can translate in the Tonal Composition that we have had the opportunity to talk and discuss between us. A good exercise could be to show a kind of reverse-painting of this subject, showing the colors, tonality, and major lines. Is obvious that the same approach could be used for any painting, also to Fauvists, for instance, but the concept here is, as we know that Cox has a great passion for the line and drawing... a good exercise would be to see whether with GIMP tools we can do an X-ray of the Cox painting.

Americo is always challenging me to see whether I can turn a theoretical understanding of color management and color mixing into something practical that can be used while making images. In particular, Americo wanted to know whether LCH really does have any advantages over HSV when exploring the color palette of a painting.

The answer is yes, LCH has a distinct advantage if your goal is to try to figure out which actual paint pigments might have been used in the original painting, because technical information about paint pigment colors is readily available as LCH values, but not as HSV values. However, if what you really want to do is make a palette from a digital image (that might or might not be a faithful reproduction of a "wet" painting, photograph, etc), then a specific answer regarding whether LCH or HSV is the better choice will have to wait for Part 2 of this tutorial. Part 1 has two goals:

  1. Provide an easy way to explore and compare the LCH and HSV blend modes
  2. Take the first step in analyzing Kenyon Cox's portrait of Saint-Gaudens, in which we learn that the Wikipedia photograph of Cox's portrait is not a very good photograph for learning anything useful about Cox's actual color palette.

Tips for following this tutorial

Reading a tutorial on digital imaging is not a very useful thing to do unless you also follow along with an example file, replicating for yourself what the tutorial is trying to show. So I would recommend downloading the Wikipedia photograph of Kenyon Cox's portrait of Saint-Gaudens. In case you are worried, it's a public domain photograph, at least in the U.S.

This tutorial uses default high bit depth GIMP-2.9. It can't be followed using GIMP-CCE because I removed the HSV blend modes from GIMP-CCE. Windows and MacIntosh users can download GIMP-2.9 from Partha's Place. Linux users can find various versions of GIMP-2.9 prepackaged in various repositories. And Carmelo_DrRaw provides Linux AppImages for GIMP-2.9.

You'll need the base layer color to follow along with this tutorial, so here's a small version of the XCF layer stack that I used to make some of the screenshots in Section B below. You can get the base layer color from the bottom layer.

Once you've downloaded the Wikipedia photograph of Kenyon Cox's portrait of Saint-Gaudens, open it with default GIMP 2.9, promote the bit depth to 32-bit floating point, and set "Image/Precision" to "Perceptual gamma (sRGB)". For exploring the LCH blend modes, make a base layer of the same color as in the small version of the XCF file, make three additional copies of the Wikipedia photograph layer, and change the blend modes and layer labels to match the small version of the XCF file. Make a similar file for exploring the HSV blend modes.

Using a Wikipedia photograph of Kenyon Cox's portrait of Saint-Gaudens to explore the LCH and HSV blend modes

Exploring the LCH and HSV blend modes

One of Americo's questions was whether the LCH and HSV color spaces are equally useful for analyzing an image's color palette, or whether it might be better to just use the LCH color space. So here is a very brief overview of LCH vs HSV:

LCH and HSV each have three components and four blend modes:

In the LCH color space, any given color is broken down into three components:

  1. LCH Lightness
  2. LCH Chroma
  3. LCH Hue

In the digital darkroom/studio, each LCH component can be used as a blend mode. There is a fourth LCH blend mode called "LCH Color", which is a combination of LCH Chroma and LCH Hue, or alternatively "What's left when you abstract away the Lightness".

Likewise in the HSV color space, any given color is broken down into three components:

  1. HSV Value
  2. HSV Saturation
  3. HSV Hue

In the digital darkroom/studio, each HSV component can be used as a blend mode. There is a fourth blend mode called "HSV Color", which is a combination of HSV Saturation and HSV Hue. After writing this tutorial I found out that the HSV Color blend mode is not available in GIMP 2.8 or GIMP 2.9. What I always assumed was "HSV Color" blend mode turns out to be HSL Color blend mode. So all images in the original version of the tutorial that involved "HSV Color" blend have been removed from this version of the tutorial.

LCH is based on measurements of how people perceive differences in color:

The XYZ color space is based on experiments to determine how the average person (A.K.A. "standard observer") actually sees colors. The LAB color space is a perceptually uniform transform of the XYZ color space, and is based on experiments to determine how people perceive differences between colors. The LCH color space is a convenient polar transform of the LAB color space, allowing to specify colors by their Lightness, Hue, and Chroma.

One practical application of LAB/LCH is quality control, allowing a quantitative answer to questions like "Does this fabric (or car paint batch or magazine print or etc) have perceptually the same colors as this other fabric (or car paint batch or magazine print or etc), or have the colors drifted far enough that people will notice the difference?" Another practical application is specifying and describing colors for paint pigments and such.

HSV is a computationally fast way to approximate how we perceive colors:

In the digital darkroom/studio, an RGB working space such as sRGB is actually a well-defined volume of colors located in the XYZ color space. The HSV color space is a mathematical transform of whatever RGB color space an image happens to be in. Quoting from the Wikipedia article on the HSV (and HSL, HSI, etc) color spaces:

While HSL, HSV, and related spaces serve well enough to, for instance, choose a single color, they ignore much of the complexity of color appearance. Essentially, they trade off perceptual relevance for computation speed, from a time in computing history (high-end 1970s graphics workstations, or mid-1990s consumer desktops) when more sophisticated models would have been too computationally expensive.

Comparing the LCH and HSV blend modes

Well, that's more than enough theory. The important thing is how useful the LCH and HSV blend modes might be in the digital darkroom/studio, and in this particular case, how useful they might be when analyzing an image's color palette.

This section of the tutorial provides "on-hover" (or "on-click" if you are using a mobile device) images showing the various LCH and HSV blend modes. Each "on-hover" image shows the result of using the various blend modes alone and in combination, and also shows a screenshot of the corresponding layer stack so you can easily replicate the results in your own digital darkroom/studio.

The first two slideshows below compare the LCH Chroma, Hue, and Color blend modes to the HSV Saturation, Hue and Color blend modes:

Wikipedia photograph of Kenyon Cox portrait: LCH Hue, Chroma, and Color
LCH Lightness + Chroma + Hue
LCH Hue over base layer
LCH Chroma over base layer
LCH Chroma+Hue over base layer
LCH Color over base layer
Base layer (the greenish bottom layer)

When hovering over/clicking on the LCH blend mode images shown above:

  • Slide 2 shows the LCH Hues for the Wikipedia photograph of Kenyon Cox's portrait of Saint-Gaudens. Notice the sprinkling of blue and green Hue splotches. I'll just tell you up front that these Hues are in the Wikipedia photograph, but almost certainly not in the original painting.
  • Slide 3 is green because the base layer is green. If I had made the base layer blue, image "c" would have been blue. And so on.
  • Slides 4 and 5 are identical: LCH Color is exactly equal to the result of combining LCH Hue and Chroma.
  • Slides 2 through 5 all have the same LCH Lightness value as the base layer. This means the LCH color space cleanly separates tonality ("Lightness") from color. You can prove this to yourself as follows:

    1. In the layer stack that you are hopefully using to follow along in this tutorial, make only the base layer and the LCH Hue blend layer visible. Then make a "New from Visible" layer. Then desaturate the "New from Visible" layer to black and white using "Colors/desaturate/desaturate/Luminance". The RGB values will be (0.566863, 0.566863, 0.566863).
    2. Now make a copy of the base layer and hide all the other layers. Then desaturate the copy of the base layer, again using "Colors/desaturate/desaturate/Luminance". The RGB values will be (0.566863, 0.566863, 0.566863). Repeat for the other layers, and results will be the same.

OK, now let's look at the corresponding HSV Hue and Saturation blend modes:

Wikipedia photograph of Kenyon Cox portrait: HSV Hue, Saturation, and Color
HSV Value + Saturation + Hue
HSV Saturation+Hue (which equals HSV Color) over base layer
HSV Saturation over base layer
HSV Hue over base layer
Base layer (the greenish bottom layer)

In the HSV blend mode images shown above:

  • The base layer (the green layer at the bottom of the layer stack) is the same color as the base layer used for the LCH layer blend modes. I just changed the layer label so you could see the HSV channel values and also compare them to the LCH channel values.
  • Looking at images "b" through "e", it's pretty obvious that HSV doesn't cleanly separate tonality from color. To prove this to yourself, try merging any combination of the HSV Hue, Color, and Satuation layers with the base layer, and then making a Luminance conversion to black and white, and compare with results from doing the same thing using the LCH Hue, Color, and Chroma layers blended with the base layer.

LCH Lightness compared to HSV Value

The two "on-hover" images below compare the LCH Lightness blend mode to the HSV Value blend mode. First up is the LCH Lightness blend mode, alone and in combination with the other LCH blend modes. All indicated blend modes are blended over the base layer:

LCH Lightness blend mode
LCH Lightness+Color
LCH Lightness+Chroma+Hue
LCH Lightness+Hue
LCH Lightness+Chroma
LCH Lightness
Base layer (the greenish bottom layer)

In the LCH blend mode images shown above:

  • Looking at image "f", at first glance it seems like the LCH Lightness layer blend mode placed over a color layer such as the base layer "a" might be a nice way to make a monotone image from a color image. But a closer examination shows that the highlights on the back of the shirt are completely flattened. I'll leave it to you to figure out why (hint: the base layer has a fairly high Chroma of 30, and combining the Lightness values of the shirt with the high Chroma value of the base layer produces out of gamut RGB channel values).
  • Looking at image "g", the LCH Lightness and Chroma layer blend modes taken together over the base layer produce a very nice monotone image without flattening the highlights. However, depending on the Hue of the base layer vs the Chroma of the colors in the original image, there is still a risk of producing out of gamut channel values.
  • Images "i" and "j" demonstrate that whether you combine LCH Hue+Chroma+Lightness, or instead combine LCH Color+Lightness, either way you get back the original image.

Next up is a look at the corresponding HSV Value blend mode, alone and combined with the other HSV blend modes. All indicated blend modes are blended over the base layer:

HSV blend modes for Kenyon Cox's portrait (Wikipedia photograph)
HSV Value+Saturation+Hue
HSV Value+Hue
HSV Value+Saturation
HSV Value
Base layer (the greenish bottom layer)

In the HSV blend mode images shown above:

  • Looking at images "f" ("Value" over the base layer) and "g" ("Value+Chroma" over the base layer), the same considerations apply as with the corresponding LCH blend modes, except in the HSV blend modes applying the HSV Value blend mode over the base layer "a" doesn't produce any out of gamut channel values. Which is a nice feature of HSV compared to LCH: Unlike working in the HSV color space, when using LCH it's very easy to produce out of gamut channel values.
  • Looking at images "i" and "j", "HSV Value plus HSV Color" doesn't produce the original image. But "HSV Value plus HSV Chroma plus HSV Hue" does produce the original image (in the LCH color space "Lightness plus Color" and "Lightness plus Chroma plus Hue" both produce the original image). Again, the HSV color space doesn't cleanly separate color from tonality. This is why using the HSV blend modes to modify an image's color or tonality sometimes produces some rather ugly artifacts.

Using the HSV and LCH color pickers to examine the color palette of Kenyon Cox's portrait of Saint-Gaudens

Having patiently worked our way through the various LCH and HSV blend modes, the next logical step is to take a close look at the LCH and HSV Hues in the Wikipedia photograph of Kenyon Cox's portrait of Saint-Gaudens. This will prove to be instructive, but the only thing we actually learn is that the Wikipedia photograph is very far from being a colorimetrically accurate reproduction of the original painting.

Yellow and orange, and also green and blue?

In the screenshot shown below:

As Americo said in our email conversation, the color palette for Kenyon Cox's portrait is predominantly in the ochre to red range. But what are we to make of the greens and blues found in the shadows (the tie, the back of the head, and below the outstretched hand) and the highlights (the back of the white shirt)? Did Cox mix two different tints of white paint on his palette for painting the back of the shirt? And did he also mix together some blue-toned and green-toned dark pigments for the shadows?

Improbable colors plus a physically impossible dynamic range

Without the benefit of Americo's comments on Cox's palette and painting style, I'm naive enough about artistic styles to think "Oh, maybe Cox decided that a dark green tie would look nice, and also maybe he painted blue and green shadows." But I found the radical Hue changes in the highlights on the shirt to be quite puzzling.

So I went back to the Wikipedia page from which I downloaded the photograph of Cox's portrait and read the camera exif data, which revealed that the photograph was taken using a Canon Point and Shoot that was set to use Auto White Balance and the Standard Picture Style. This means the camera itself:

  • Stretched the image's dynamic range.
  • Applied a heavy "S-Curve" to the image to darken the shadows and brighten the highlights.
  • Added saturation to make the colors more pleasing.

To further complicate the matter, we have no way to know how far off the Auto White Balance might have been. And we don't know what post-processing the photographer might have given the camera jpeg before uploading it to Wikipedia.

What we do know is that no physical painting made with real paints on a real canvas has a dynamic range that extends from solid black to solid white. And looking more closely at the photograph's tonal range — the "marching ant lines" in the image below indicate the portion of the image used to make the displayed histogram — indeed portions of the photograph are solid white and portions are solid black:

So the bad news is that unfortunately we haven't learned very much about Kenyon Cox's actual color palette. But we have excellent reasons to suspect that the original painting is considerably less colorful than the photograph, and we can attribute the technicolor Hue splotches in the shadows and highlights to artifacts created by clipped and near-clipped channel values in the Wikipedia photograph. And perhaps we've learned to be suspicious of colors in photographs of artwork.

A more colorimetrically accurate reproduction of Cox's portrait of Saint-Gaudens

It would be nice to find another and hopefully more colorimetrically accurate reproduction of the original portrait. Which I did. Kenyon Cox's actual portrait of Saint-Gaudens currently belongs to the Metropolitan Museum of Art, which supplies a digital version of the painting that is accompanied by an OASC icon that clearly identifies the painting as both public domain and part of their Open Access for Scholarly Content, and also provides a link to a high resolution downloadable digitized version of the painting:

The Met Museum's digitized version of Kenyon Cox's portrait
For easy comparison, the Wikipedia photograph of Kenyon Cox's portrait

Well, the Met Museum reproduction of Kenyon Cox's portrait of Saint-Gaudens is certainly a lot less colorful than the Wikipedia photograph! And speaking frankly, I somewhat prefer the Wikipedia photograph. Sigh. For what it's worth I never claimed to have refined artistic sensibilities.

Summary and on to Part 2

Section B of this tutorial compares and contrasts the LCH and HSV blend modes:

  • LCH is a physically meaningful color space that is based on how we perceive colors and color differences. HSV is a computationally fast way to approximate how we perceive colors.
  • LCH cleanly separates Color from Tonality, and the LCH Color blend mode is exactly equal to the LCH Hue blend mode combined with the LCH Chroma blend mode. HSV doesn't cleanly separate Color from Value, and the HSV Color blend mode is not equal to the HSV Hue blend mode combined with the HSV Saturation blend mode.
  • On the one hand, in the digital darkroom/studio LCH color blend modes can sometimes produce out of gamut RGB channel values. On the other hand, although this tutorial didn't show examples of artifacts from using the HSV blend modes, such artifacts are a result of the fact that HSV doesn't separate Color from Tonality.

Section C evaluates the Wikipedia photograph of Cox's portrait of Saint-Gaudens in terms of whether it's likely to be a colorimetrically accurate reproduction of the original painting, in the course of which we:

  • Examined the results of applying the Wikipedia photograph of Cox's portrait over a solid green (actually cyan) base layer using the LCH and HSV "Hue" blend modes.

    Whether looking at the LCH or the HSV Hue blend mode, the blue and green splotches — which have nothing at all to do with the color of the base layer — please experiment for yourself to confirm! — made it was pretty obvious that "something wasn't right" about the Wikipedia photograph of Cox's portrait of Saint Augustine: Painters who use predominantly ochre and earth-red palettes don't usually use blue and green pigments to produce cool-hued shadows and highlights.

  • Examined the dynamic range of the Wikipedia photograph of Cox's portrait, and concluded that the photograph has a dynamic range that simply can't be produced using physical mediums such as paint on canvas.
  • Took a quick look at the Metropolitan Museum digitized version of the same painting, which certainly is much less colorful than the Wikipedia photograph, and hopefully is a more colorimetrically accurate representation of the original painting.

Part 2 of this tutorial (not yet written) will explore the dynamic range, tonal distribution, and color palette of the Met Museum's digitized version of Cox's painting. Along the way, I'll try to provide some good answers to Americo's question of whether LCH is more useful than HSV for determining a painting's color palette.