Histogram stretching and color reproduction

0
834

I am continually surprised that 12 years after Michael Reichmann first spoke about the concept of “Expose to the right,” translated as “Histogram Righting,” and addressed by the controversial José María Mellado in his popular book “High-Quality Digital Photography,” there are still people writing and talking about the topic as one of the great premises of digital capture.

Let us remember that the concept of Histogram Righting was related to an optimization of noise in the digital image, since the signal-to-noise ratio was improved through overexposure and a subsequent equivalent underexposure during development.

Although noise optimization and tone mapping have been widely discussed online, perhaps the consequences for color reproduction have been less addressed.

Tone Mapping

The righting technique forces us, on the one hand, to overexpose without clipping the highlights, say for example +1EV, and then during development to underexpose by the equivalent amount, that is, -1EV. To perform this underexposure, the user usually uses the Adobe Camera Raw or Lightroom slider called “Exposure.”

To carry out this experiment, I calculated the sensitometric curve of a camera (Nikon D90) to understand the response of the highlights, midtones, and shadows, using the densitometric scale of my Colorchecker as a reference. In this way, we can perfectly see the behavior of this camera at different exposure values.

derecheo del histograma, mapeo de tonosAfter obtaining the sensitometric curve of this camera, starting from an image, I performed an underexposure using the “Exposure” tool in ACR, which returns curves quite close to the different exposure steps of my camera. In the image on the left, the lines labeled “-PS” correspond to the adjustment made with PS, and those labeled “-Cmra” are the camera’s sensitometric curve. The three lines B, G, and N correspond to the highlights, midtones, and shadows, respectively. The correction values in ACR are shown at the top of the graph.

Therefore, we conclude that underexposure during development leaves the image at a point very similar to if we had exposed correctly. That is, at the achromatic level, an image can be corrected during development and respect the tone transfer between reality and the image. This is essentially what we pursue with tonal reproduction adjustments using the curves tool during development.

Color Reproduction

However, assuming that tonal reproduction is ensured, we face another problem, which is color reproduction. When we underexpose or overexpose, initially we act on the achromatic dimension of color, that is, the luminosity or lightness (L*).

However, the chromatic dimension of color, that is, the Hue or the a*b* attributes, does not vary in the same way. That is, if we underexpose, L* decreases; if we overexpose during development, this attribute is recovered. The problem is that during underexposure or overexposure, the chromatic attributes also vary; this is known as “hue shifts.”

derecheo histograma reproducción del colorHue shifts are a common problem during raw development; in fact, it is a problem considered during color management during development with more or less success, with the difficulty that some colors will suffer more or less pronounced hue changes depending on the level of under- or overexposure. In the image on the left, we can see the difference in hue of each color sample after the exposure adjustment.

Thus, for example, I took two shots of a Colorchecker with a +1EV difference. After underexposing by 1EV during development and trying to match the tonal reproduction between them as closely as possible, we get the following delta report:

Patch dE* dL* da* db* dCh
1 1,06 -0,44 0,38 0,89 0,96
2 9,62 1,49 -8,86 3,43 9,5
3 2,87 -1,65 -0,39 -2,31 2,35
4 0,57 0,07 0,3 -0,48 0,56
5 1,18 0,19 -1,15 0,17 1,17
6 4,29 -1,24 -2,06 -3,56 4,11
7 0,25 0,23 -0,06 0,08 0,1
8 0,33 0,32 0,07 -0,08 0,11
9 1,84 -0,09 -1,55 -0,98 1,83
10 1,15 -0,61 0,67 0,71 0,98
11 1,77 -0,28 -1,69 -0,45 1,75
12 0,36 0,24 0,22 -0,16 0,27
13 0,48 -0,05 -0,31 0,37 0,48
14 0,69 -0,54 0,41 -0,14 0,43
15 6,5 1,77 -6,15 1,13 6,26
16 0,2 0,14 0,09 0,11 0,14
17 1,05 0,15 -0,36 -0,98 1,04
18 0,32 0,24 -0,02 0,21 0,21
19 2,42 -0,19 2,28 -0,79 2,41
20 0,16 -0,11 -0,11 0,04 0,12
21 0,55 0,03 -0,04 -0,55 0,55
22 9,31 2,36 -5,03 7,46 9
23 0,94 0,75 -0,18 -0,52 0,55
24 0,48 0,18 -0,31 0,33 0,45

As we can see in the table above, while most color samples show very low delta values (< 2), there are specific samples with dE* values of 6 or 9, which are especially high values implying a considerable hue change. If we look closely, we can also observe that the dL* error is essentially low; this is because both images were approximated as much as possible at the tonal reproduction level. However, we observe that most of the error accumulates in the chromatic part dCh, da*, and db*.

Conclusions

Besides discussing a technique that dates back to the dawn of modern digital photography, the performance of modern cameras is far from the models of more than 10 years ago.

Histogram righting is nothing more than an exposure error, an error that, although it can be corrected at the tonal reproduction level during development, we find that histogram righting can cause serious color reproduction problems, since there are colors that cannot be recovered by a mere exposure adjustment. This is one of the problems that most limit us from achieving good levels of color reproduction during raw development, as it is common to start from images with exposure errors greater than 1EV.

Therefore, in an era where modern DSLR models have considerably low noise levels, continuing to sacrifice good exposure practices in favor of a supposed improvement in the signal-to-noise ratio is, in my opinion, a completely absurd habit.

Unless we seek creative effects, while we seek fidelity or the maximum correspondence between digital copy and scene, proper exposure is as necessary as focusing on the correct area.

However, not all is lost; review this other article on how to right the histogram without altering color “How to right the histogram without altering color“