We’re increasingly demanding more and more demands from our cameras. We already have 14 steps of dynamic range, very low noise at 3200 ISO, an extremely fast autofocus and compact lightweight bodies, but… what about color? Manufacturers don’t say anything remarkable about the color that is captured and how their cameras reproduce it, beyond the fact that they can produce “perfect” and “wonderful” colors (who would dare to doubt it!).
Well, many current DSLR cameras already have the ability to record a tremendous color gamut, reaching and even slightly exceeding the ProPhoto space in some cases. This makes them truly capable color registration tools. So, why don’t we adjust that wide color gamut with advanced color charts? Many photographers use the ColorChecker, introduced in 1976, with only 24 color patches which does not seem to be the best option. Firstly, because it was not designed for that purpose, but to verify the results.
In this article I want to explain my experience calibrating cameras over the last 17 years. After having calibrated dozens and dozens for clients (like ZARA HOME, MANGO, DESIGUAL, for some art galleries or the Spanish National Biblioteque to name a few) with almost all the existing color targets, I realized that the results could be improved, but there was no color chart on the market to achieve this, so I decided to design it myself. Or at least try.
The HR-1 SuperChroma chart
The HR-1 SuperChroma is a color chart that has been designed with the aim of producing the best possible camera profiles to date. Unlike other charts, the SuperChroma breaks the mold and offers innovative solutions. It is the result of 3 years of work and has been created based on several fundamental premises:
1. To offer the largest gamut available today, thanks to giclée printing with 12 pigmented inks.
2. To have the most complete set of patches for profiling capture devices currently, with 999 patches.
3. Achieve the most perceptually homogeneous color patch distribution ever created.
4. Provide the greatest number of visual aids.
Most charts on the market don’t have enough color samples to profile a camera. And, no, 15 or 20 patches are not enough to accurately profile a digital camera for EVERY possible color.
Let’s assume that a camera produces 16.7 million colors. According to this, and by a basic rule of three, from that total 1 or 2 million would be reds, right? So how do you adjust a million different shades of red with a card that has just 2 reds? No doubt it would be a complicated task. It’s true that a card with 200, 500, or even 1000 patches would still be small, but it will be more capable because it would at least have 15, 20, or 50 red patches.
In the following chart you can see the difference in number of patches among the most popular ones in the market:
Those that do have a sufficient number of color patches are made of chemical paper, limiting their color range substantially, but that is not the only problem.
This graph shows the gamut of a Colorchecker, a chart printed on chemical paper (like the IT8 or the Hutch) and the SuperChroma:
However, the big difference of the HR-1 SuperChroma is in the distribution of its patches in the color space. Instead of following the classic gradient pattern, it distributes all its patches perceptually evenly across the color space.
On charts like the IT8 or the Hutch, which were created based on the concept of gradients, there is a lot of patch density in certain areas (especially near grayscales) and very little or no patching between the extremes of maximum gradient saturation. This makes the patch distribution uneven and consequently worse for a color profiling program.
In the following graph you can see the patch distribution of the Hutch chart, where you can clearly see that there are redundant patches in the central area and, in the outer area, empty areas with almost no information.
In the HR-1 all patches are evenly distributed, even in black or white, avoiding wasting patches in those areas, which produce redundant information.
This allows, in short, that when creating profiles, the software has information throughout the gamut and there are no empty areas as happens to other charts, which causes some colors to be well profiled and others not. In other words: constant distribution = constant quality.
Theories are all very well, but the photographer is usually interested in practice.
Thanks to the fact that the SuperChroma has had a relatively long period of testing as a prototype and that some of these have been in several of the largest textile companies in many countries, such as Zara Home or Mango, I have been able to test the results in real environments of maximum difficulty and with the highest levels of demand.
The HR-1 SuperChroma has been designed mainly for use in e-commerce and artwork reproduction. Therefore it can produce excellent results in a scene-referred workflow and also in an output-referred too.
Problems with the reds are very common, and photographers know this well. Here’s a real example of a red shirt in an online store:
So far so good: one expects to receive a dark red shirt and not too saturated, but the problem is that the shirt that arrives in a bag is this one:
To check the color, I measured the garment with my EyeOne Pro2:
I could see that the color turned out to be 181 | 12 | 42 (in sRGB space). This contrasts with the values that can be measured in the online shop, which are: 149 | 35 | 34. In my capture you can see that my measurement matches the RGB values in several points of the garment (obviously not in all of it, as it depends on the folds and the lighting):
Therefore between the two colors there is a not an insignificant difference in the tone of 15.7 (Delta ab, which measures only the change in color and not in the brightness, like Delta E)
This case shows how well SuperChroma can behave, but the really interesting part comes when it is tested with some of the most difficult colors one can find. For example, the beige colors:
When in fact the actual garment is like this:
This last image has been rendered as scene-referred in Capture One Pro, so the skin may look not so pleasant, but the garment’s tone is much more accurate:
This last image has been rendered as scene-referred in Capture One Pro (using the Superchroma profile plus linear curve), so the skin may look not so pleasant, but the garment’s tone is much more accurate.
Even worse than beige is highly saturated magenta, which almost always drifts into redness:
When the real result is this:
But the most extreme examples are the phosphorite colors. Many photographers get nervous just thinking that their client wants to photograph a garment in phosphor yellow, like this one:
This is the original RAW developed with the generic (C1’s embedded) profile:
And this is the result with the profile generated with the HR-1 SuperChroma:
It should be noted that there are no adjustments of any kind in the development, except for the WB and the color profile.
Results and data
I think the results clearly show what HR-1 SuperChroma is capable of. But… how do you know if the results are correct?
The way to check is to analyze the color of each of the patches in the color chart capture once it has been developed with the same ICC profile created for the camera. Let’s see it.
The first thing I have to say is that to make a reliable analysis it is important to point out that you should not use the same chart used for profiling. If I use a Colorchecker to profile the camera and then I use the same chart to check what the results are, I’ll get values that are too good. This is because they are used to check the same colors that the profile uses internally to adjust the color response of the camera. By doing so you get false good values, because the colors used for profiling are precisely the ones that will give the best results. That is why it is important to use different charts, to obtain real-world values.
The graph represents the chart itself, in 3D view, so that the height of each column is the color deviation from the reference. The greater height, the greater deviation. This deviation is measured with the Delta E value, and if it is 1 or less, it is invisible to the human eye. From 1 to higher values it becomes progressively more visible.
In the table of values below the 3D graph you can see several statistical values related to the mentioned Delta E. They are important values because they summarize in a few numbers the quality (fidelity to the original) of a capture:
The most important is the one in the lower left corner, “ΔE00”, which stands for Delta E 2000. A value of 2.9 is not bad, and indicates that on average there is a clearly visible difference between the reproduced colors and the originals.
In this other graph the color difference is much more clearly seen:
The problem comes when we analyze the fidelity of an IT8 chart captured and developed using the profile created with a ColorChecker:
Where you can see that the values are now much worse (and real at the same time):
The reality is that on average you get a Delta E of 3.8 and a maximum of 7.4,with a Delta E ab of 18. Not very bad, but not suitable for high-end results.
What values would be obtained when capturing an IT8 using a profile created with the IT8 itself? Let’s see it: in this case I have used an EGM Wide-gamut IT8)
In this case the values are much better (lower), going down to a Delta E 1.60 on average, a really low value.
But, again, these are false low values because of using the same chart for profiling and checking. Now let’s see what results we get by profiling with a SuperChroma and checking with an IT8:
Result: an average Delta E of only 1.78. And now these are real values.
It should be noted that using median would better than using mean, to avoid errors caused by out-of-gamut colors that doesn’t let us see the real deviation.
The highest peaks correspond to colors that are out of range of the camera space, so they will always give high errors. This means that with a SuperChroma you can achieve what’s probably best fidelity in the industry up-to-date.
In this image you can visually check how minimal the deviations are.
It’s true, perfection is impossible (at least, for now), but the resemblance is so great that the differences are close to the imperceptible.
As a comparison let’s see the result of the Colorchecker’s profile on this same IT8:
It’s worth remembering that in this case the average was Delta E 3.8. It is also important to point out that the differences are almost exclusively in the color and not in the grey scale, which confirms that there are no problems in the exposure or the WB in the capture. It is purely a color management problem.
If you are interested in knowing more about this chart or purchasing it, you can find more information at: http://www.superchromachart.com/
Thanks to Danny Pascale, CEO of Babelcolor, for his invaluable help. To Jose A. Fernandez of EGM for his help with 3D skin tone mesh optimization. To Francisco Colon for his help correcting this text. To Juanjo Agirreolea and Matt ‘Techmoan’ for their help with some final aspects of the chart. To my family as well and some friends and people I know who let me take skin color readings from various points on their body for the skin patches. To Enric Galve for his support from EGM Laboratoris Color.