Perceptual and relative colorimetric rendering intent

Colorimetric rendering intents : relative or perceptual ?

When you do a conversion in Photoshop in particular, you must therefore :

• First, make sure that the original image has an ICC profile,
• then, choose the destination ICC profile - your printer and paper -
• But also, as we have just seen, choose the rendering intent.

What is it about ? Simply the way out-of-gamut colors are managed during a conversion. Do we "delete" the colours out of gamut - logical because they are not printable ? Problem : it's destructive. Are they replaced by other colours ? Which ones ? What about the colours used for replacement ?

Let's see it now.... Depending on the rendering mode chosen, the conversion rules will change. The most suitable for photography is very often, in my opinion, the relative rendering intent but the most "logical" because so-called photography is the perceptual rendering intent. I will describe them now.

Let's take an example to feel the problem : let's imagine that I want to print an image. It has an independent color space - red line on the illustration opposite - and I know that my printer's color space is much smaller for these green/yellow colors - white line -. Colors from my original, around point A, should be non-printable - the associated L*a*b* color is not printable. I want to try to print these colors anyway so I have to convert my image in another space, the destination space - inside the white line - to translate them into the right RGB values BUT HERE THE GOOD RGB VALUES DO NOT EXIST BECAUSE THEY ARE NOT PRINTABLE.
In addition, point B is common to both spaces, so the associated L*a*b* color belongs to both spaces. This L*a*b* color can therefore be converted to other RGB values.
Under these conditions, how can I still print these yellows / greens of my image that are outside the reproducible space of the printer so as not to completely distort my image once printed ? What does the out-of-range color printer do ? Does it remove them ? In addition, on my original image, the distance between A and B gives me a certain visual "sensation"; how can I keep it after the conversion ?

But this raises several problems :

1 - 1 - If the out-of-space greens, which are normally not printable, are placed in the printer's space - inside the white line - many colours will, in fact, be "superimposed". One might think then that the only choice available to us is to keep only color B and eliminate A. In other words: what is the decision to keep and eliminate ? Do we really eliminate it ?
2 - Moreover, if I put (replace) the color A on the color B and the colors of the red line on those of the white line, it will be impossible to differentiate them on the print while they were different in the original space.
3 - Finally, I would like to remind you that the yellows of point A were at a certain distance from the yellows of point B in my original image, which gave me a certain visual sensation.

To solve these problems, mathematical researchers have therefore invented conversion rules called rendering intents : the perceptual and relative rendering intents for photographers.

In perceptual colorimetric rendering intent

The ACE Photoshop conversion engine will work a lot ! It will indeed "bring" the greens out of space so as not to completely remove these nuances and preserve my visual impression, my visual perception. To achieve a conversion that respects the visual sensations of the original image, Photoshop's ACE conversion engine will proceed in two steps. First of all, it will "place" the colors of the red line (boundary of the image space) on the white line (boundary of the printer space) because the colors of the white line are the closest visually to the colors of the red line. The colors of the red line are the most saturated colors of my original and the colors of the white line are the most saturated colors of my printer. The most saturated colors of my original are preserved and replaced by less saturated colors.
Then, to maintain a visual sensation close to the original sensations, it will slightly and gradually "shift" the colors inside the printer space as shown in the figure above. B will move to B' and so on... B's place being "liberated", he will be able to put a color out of gamut in its place. So it's a progressive game of musical chairs! As he could not do this indefinitely otherwise colors would disappear from the "other side", therefore towards the reds, he makes "disappear" some nuances of the original image in the gamut as he "pushes" the colors inwards but while trying to keep the same perception, hence the name of this mode. The image will lose some of its original saturation but the visual sensation will remain very close on the print.

In brief :

• The red line goes on the white line = the most saturated colors of the image are printed with the most saturated colors of the printer; There is a loss of saturation on some colors.
• The A color "disappears" because it cannot be printed = it is replaced by the A' color, the closest but printable color. At the same time, A' is replaced by A''. From time to time two different colors in the original space are replaced by the same color but this time it can be printed.
• The relative distance between A and B is replaced by a smaller distance A' and B' so the "spirit" of the original photo is preserved as best as possible. This is why an image must be processed as long as possible in its original space. It's the only way not to lose information.

Conversions in this mode are destructive since all colors are modified, even those that were printable. They must be limited and done at the last minute. Finally, the converted image will appear slightly desaturated and only slightly less contrasted if the printer space is really too small for these colors. It is then sufficient to apply a level or contrast layer to restore - almost - the original L*a*b* colors.

In relative colorimetric rendering intent

The ACE conversion engine of Photoshop works much less because it is then a question of simply "eliminating" all the original colors out of gamut. Did you say eliminate ? obviously not quite ! In fact, all the colors out of gamut are replaced by a single color, the most saturated color that the printer is able to print. All colors between the red line and the white line are replaced by the only colors of the white line. Phew ! On the other hand, the L*a*b* colours common to both spaces remain unchanged.

Summary :  Thus, if the out-of-gamut colors "disappear" or rather are replaced by the most saturated colors that the printer is able to print, it is necessarily at the expense of the loss of all the nuances of the most saturated colors on the original image. The original drape partially disappears as many shades disappear. On the other hand, the common colors are not modified at all as in the perceptual rendering intent. This mode is therefore perfectly suitable when the original image contains very few out-of-gamut colors or when an original color is printable and must not move at all.

How to choose between perceptual and relative colorimetric ?

The most important thing is that there is not a better way of rendering than the other, it depends on the photo to convert and more precisely of Lab colors contained in the file.

Reminder: In nature, the colors are not always very saturated and that is why the choice of working only in sRGB is not "catastrophic". While it is fashionable to work in larger color spaces to show that you know a lot about color management, I find even smarter to stay in sRGB in most cases because the sRGB color space already contains enough colors, to be only rarely forced to work in a larger color space - but it all depends on what you photograph -. Each picture we take does not contain millions of colours and not all photographers take pictures of turquoise blue seas, designer dresses with intense reds with flash or even leaves in the sun in proxi photo !
Thus, if your original does not contain any or only a few colours out of gamut for your printer, they will not be "lost" by a relative colorimetric rendering intent ! It is then the most judicious because it does not transform - and therefore does not deform - the colors within the common gamut. As if by chance, it is the one that works best most often and is perfectly suitable ! Even if you work in ProPhoto RGB, it is amazing how well this rendering mode often works very well. Your image did not necessarily contain many out of gamut colors for the printer whose gamut is much smaller.
If your original contains non-printable colors in significant quantities, it may be wise to use the perceptual rendering intent. Certainly it changes all the values of your file but it preserves much better some materials, some drapes or gradients, some relative distances between two colors as explained above.

How does it get actually materialized ?

Where the relative rendering could transform non-printable colours into a solid without shades since all the out-of-gamut colours are replaced by a single colour, the closest printable colour, the perceptual mode tries to keep these shades, even if it is obliged to replace them by others, very close visually but less saturated. I will try to show you, I hope, with the illustrations below.

Now that my image is correctly displayed and in a neutral and large working space, I can touch it up as I wish - possibly with a eyedropper -.

In fact, each device, each image etc.... has its own ICC profile and can be interpreted by software such as Photoshop. It can assign a profile to an image or "translate" the colors, we say convert an image, from one device to another. The devices can therefore communicate via the CMM - color conversion engine - and their ICC profile. In Photoshop since version 6, it is called: ACE color engine. There wasn't any before this version in Photoshop (so it was a long time ago).

To be able to work, the CMM needs to know which ICC profile is assigned to an image - source ICC profile - to know which colours L*a*b* it has to do for given RGB signals and to which device to send it - destination ICC profile -, thus converting it into R'V' B' signals.

The CMM is the central element and is based on the L*a*b* colours (the universal standard) and not on RGB or CMYK signals because, as we have seen several times, the L*a*b* colours are absolute. It knows it must transmit as information such or such color L*a*b* (therefore absolute) and not such or such RGB value (relative) thanks to the ICC profiles and knows to which RGB or CMYK signal it corresponds for this one and only this one. It will therefore translate this L*a*b* color, a different RGB value for each device - even if they are obviously close enough - into another signal R'V' B' or C' M' J' N' so that the destination device reproduces the same L*a*b* color. And the strongest thing is that if it cannot do it directly, it will replace it with another one without betraying the perceived visual sensation. This is especially his great strength ! This "translation" operation is called a conversion and there are four different ways to do it according to the desired rendering. We will discuss this in detail on the next page. 

When does a photo need to be converted ?

So a image need to be converted at least once in its lifetime to match what our Photoshop eyedropper measures and what we see. Thereafter, we will automatically make or undergo further conversions to modify the RGB values of our photo - so always without changing the colors or as little as possible - to broadcast this photo on the Internet (sRGB color space), to be displayed correctly on our calibrated monitor (it is done on the fly in the graphic card) and therefore take into account its own display defects, or even each time it is printed.

Important ! When converting to a neutral color space, you should choose it a little larger than the ICC profile of the device you used. That's why there are many of them, from the smallest sRGB to the largest ProPhoto RGB. 

How is the conversion done ?

The paragraph above has just seen the different roles of conversion. When you choose to do a conversion, what really happens to the RGB values of my image? How can the conversion tool, the CMM, best preserve the L*a*b* colors, i.e. the visual sensation of the photo if the destination device does not allow it, as is unfortunately often the case with printers ? The conversion being in a way a translation whose mission is not to distort the original...

More or less large colour spaces...

As we have seen on the page dedicated to color spaces, the colorimetric spaces called peripheral or work spaces, are more or less large. Some are so large that they include all the others (ProPhoto RGB or some digital cameras). Some are so small that they are encompassed by all the others like some printers on matte paper. But sometimes, however, some are a little larger in one colour or another and vice versa for another. This can be seen in the illustration on the right. Overall, the color space contained in the ICC profile of this printer/paper pair is smaller than the gamut of the screen except towards the blue-green ones. These colors are then printable but invisible on the screen (not displayable as they are, therefore displayed but less saturated than their original saturation). More often than not, the opposite is true.

In other words, some colors may be contained in the original file but not printable (or even displayable). So we observe two scenarios :

• Either the original colors are printable (in the destination gamut)
• Either the original colors are printable (in the destination gamut)

In the first case, if the colors of the photo are printable, the conversion is only used to change the RGB values of the photo to obtain the same printed color on the print. So far, so good! But what if the second case or if the gamut of the image is much larger than the gamut of the printer or... the opposite ?