Study of the highlight clipping using RawDigger histogram
It is useful to know how to determine practical clipping point for at least two reasons: the most common is that we need to know clipping point to determine the headroom in highlights and tune the camera settings to provide more reliable overexposure warning (“blinkies”); another one is dynamic range analysis (signal-to-noise ratio, SNR), and measurements of utilized well capacity (full well capacity, FWC, is seldom used in modern cameras).
Different cameras, even if based on the same sensor, may render extreme highlights around the clipping point differently; and differently, even with different values of clipping points, depending on ISO setting. It is important to recognize the look and calculate the practical clipping point, which is not always the same as the maximum raw value. Here we will try to demonstrate the typical “looks” of the histogram of the clipping zone.
Any curved metal or glass surface is a good source of specular highlights. In our case, it is what we want as we will be analyzing how the clipping exhibits itself on the shots like these from Imaging Resource:
A shiny metal ball, or any metal bowl, is also a good target to inspect highlights and clipping. Having dark background behind the shiny surface helps to isolate highlights on the histogram. If you are anticipating performing dynamic range analysis having a black trap in the same shot saves time (you can make a black trap yourself, see http://www.imatest.com/docs/veilingglare/ #target, or use the Datacolor SpyderCUBE, as we did below).
We are going to open these shots, made by different cameras, in RawDigger. OvExp checkmark in the Display section causes a red overlay to appear over the blown highlights, similar to the “blinkies” in-camera overexposure indication. To make these small areas more noticeable we additionally placed red rectangles over these areas.
… and look at their histograms.
To look at histograms’ structures in more detail, let’s switch on the “Log scale” on the Y-axis, and for starters turn on “Linear X-Axis” and check Auto both for X-axis and Y-axis range:
Upd: Francisco Disilvestro made an important point:
We'd like to add to check that you have the "Masked Pixels" option unchecked in the Preferences -> Display options.
Some cameras have reference pixels hard-wired to min an max numeric levels in the masked pixels area, which will show as a spike in the histogram where there should be none. This is easy to note in cameras that have different blown out levels for different channels, such as some Nikon models. In those cases, you will see a spike at blown out level and then another small spike at the maximum numerical value (depends on bit-depth)
For even more detailed view we reduce the displayed range along X-axis so that it will include, more or less, only the area of highlights. Switching off “Auto” in the “Linear X-Axis” section, we can set the left margin of the range close to the structure in the highlights. Now we can set the bin size to 1. We will be referring to this set of actions as to zoom in to the highlight portion of the histogram.
For overexposed shots with blown-out highlights the extreme right of the histogram usually takes one of following typical shapes:
- a two-slope structure, which resembles the Empire State Building,
- a two-slope structure, resembling a bell,
- a one-slope structure, which brings to mind a wave hitting the wall,
- a spike,
- a hybrid structure (may be different in every color channel)
1. The “Empire State Building” as displayed in log scale Y-axis mode (a typical example would be SONY cRAW/ARW2):
2. A “bell” as displayed in log scale Y-axis mode (Canon):
For those wondering why we classified this as a bell-shaped rather than an the “Empire State”, this is why:
3. A “wave hitting the wall” as displayed in log scale Y-axis mode (Sigma):
4. A “spike” as displayed in log scale Y-axis mode (Pentax, Leica, Samsung)
5. A hybrid structure as displayed in log scale Y-axis mode (a typical example would be shots from Fujifilm)
Some cameras, such as Canon and a few Panasonic cameras have a certain peculiarity – their maximum (that is a clipping point) changes depending on the ISO setting.
Take for instance this particular histogram of a shot from a Panasonic GM1 at ISO 125:
This seems normal, in and of itself, but, if one were to take a look at the shot at ISO 400…
…It becomes plainly evident that the maximum has shifted at least 500 levels to the right, and the bell-shaped curve is no longer there.
n certain cases, the «blown-out highlight» histogram’s structures will not be aligned vertically, meaning that the center values of the structures will be different for different channels.
Most often, this is the indication of color channels (white balance) preconditioning, like the one Nikon is using on of their cameras since D2X (D5300 and D3300 are two recent exceptions).
This is as it looks in full-range, and this is how it looks if we zoom in closer. We see that maximums are different for different channels:
Cameras where color channel preconditioning is used are recognizable by the regular one-pixel gaps in the histograms of red and blue channels (sometimes, additionally, they have different clipping points for red, blue, and both green channels, as on fig.24). This is because of how the preconditioning is implemented – it is digital multiplication of data in red and blue channels by a number slightly higher than 1, applied before the data is written into a raw file.
Which value should be considered as practical maximum?
The conservative approach is to consider the value that starts the overexposure “structure” to be the maximum, as the higher values are mostly pattern and processing noise. For “hit-the-wall” and “spike” types of highlight structures, it is the spike itself. However for something like we have on fig.22, which is not a pure “hit-the-wall”, the clipping starts at 3935.
So, if we return to the histogram shown on figs. 9 and 10 for Sony A7R, we will see that here the peak value is 15860, while the bulk of the structure starts at 15635.
Now we can go into Preferences and, following the conservative approach, enter this second number to Overexposure Detection section in Manual level all channels field.
Note: If your camera is one of those with white balance pre-conditioning, you will need to enter per-channel values in Manual per channel section R, G, B, G2 fields
Now we can hit Apply and inspect the main view. If you have done everything above correctly, you will have a very accurate display of the overexposed areas.
On a normally exposed image, you will have at least a few of the overexposed pixels in every major specular highlight (if there are such highlights in the scene, of course). If some highlights that are specular in nature do not have overexposed pixels even if looking at 100% magnification the image should normally be considered as underexposed. As it was already mentioned, curved metal and glass surfaces are the first candidates for these specular highlights, as well as all of the sources of light in the image.
And two final notes to those of you who decided to experiment with your cameras:
- It is easier to recognize and analyze the specular highlights if the area is large enough and placed over a much darker background.
- If you are using a full-frame camera, or a medium format camera, place the specular highlights slightly off-center to avoid the areas of technological stitches to have a better view of the histogram (different parts of the stitched sensor often have different characteristics and the histogram is not so easy to read if those are mixed.)
To see how the sensor is stitched lets take a fully blown out shot:
To exaggerate the contrast we use the Per Channel black level settings in Preferences:
… and inspect the displayed structure:
Though the difference between the stitched halves is very small it is obvious that the “left” part of the sensor is more uniform and better suitable for analysis.
This article in PDF format: RawDigger_Histogram_Part_III_Overexposure_Shapes.pdf
FastRawViewer is a must have; it's all you need for extremely fast and reliable culling, direct presentation, as well as for speeding up of the conversion stage of any amounts of any RAW images of every format.
Now with Grid Mode View, Select/Deselect and Multiple Files operations, Screen Sharpening, Highlight Inspection and more.