[GUB]

It is late Autumn here – too late for any roses so I had had to resort to using a particularly bright red set of undies freshly bought to run a few experiments to find out the best way to present red tones in all their glory.
The first experiment is to try and see how to identify the clipping point of colours in the histogram and to get an idea of how much margin of error is in the on camera histogram.
Camera Pentax K-1
So the day is very cloudy with a very consistent light.
The camera is in manual mode.
WB Daylight
CI Natural
Hi Dynamic Range =off
I set the first exposure for the red Histogram to be just clear of the right handside.(Which is where I would normally place it.) First image
Then with each shot I increase the exposure by 1/3 of a stop.
After downloading into darktable I bring the exposure back accordingly so the Histograms can be compared.

So note the EV settings in each screenshot.
I can identify clipping in the 4th image (1\30 sec) and the spiking in the 5th shot is an absolute giveaway that clipping is well under way.
The clue in the 4this the front of the histogram is slightly more sawn off and the histo has moved slightly left.
So on that basis if I set exposure so there is a visible gap between the histogram and the Right side then I have a maximum of one stop of margin of error.
But that is on this individual example and it may vary with say a more contrasty or less contrasty image.

Attached Images

   

---

Last edited by GUB; 05-21-2019 at 03:19 AM.

[GUB]

The next test is to try and see if varying the jpg settings change the displayed histogram.
I chose the three settings that I suspect may change things – Dynamic Range highlight correction, Custom Image, and White Balance.
I did not change the exposure settings of the camera.
I only had “auto”as a choice for highlight correction and it may be the exposure didn’t kick the correction in.
No visible change from changing Custom Image from Natural to bright.
But changing WB from daylight to cloudy shifted the histogram about 1/4 of a stop. But this is probably because the image was red dominant so warming it would have a big effect.
My conclusion is the settings do not create a significant change in the margin of error as tested in the first post above.

[GUB]

The next test was to check out something I had suspected and it has proved to be a biggie for those trusting their camera meter especially those in “spot”mode.
Now we all know that the meters are tricked by black cats and snow. It has been that way since in-camera metering started.
So lets have a look at a colour wheel of fully saturated primary and secondary colours.These colours have to be treated like they are pure white – the slightest increase in exposure will clip them.
But the camera meter doesn’t give a hoot what colour they are but just cares about how much light is falling on the sensor.
So I was suspecting the meter was essentially seeing a greyscale of the colours.
Now have a look at the colour wheel once it is greyscaled – the reds and blues have ended up with a tone that is darker than the 18% grey that meters tend to calibrate to.
Could it be that the meter is metering red (and blue) this way and so giving a totally different exposure to them compared to how it would meter white.
So in this test I have set the camera to AV mode and let it choose the shutter speed.To state the obvious the incident light falling on these subjects remained constant so the correct exposure for one should be the same for all. But we also already know the meter will tend to over expose the black and unde rexpose the white.
Just to clarify, these are clothes on the clothesline plus a sheet of photocopy paper for white. The black woollen shirt would be a very effective black with no reflection.
I am pretty sure my very rough estimate of 18% grey has proven fairly accurate in the towel and so anyone understanding incidence metering would recognise that its exposure (1/50 sec) would then be correct for all subjects.
In conclusion, if I have got this right then it is fair to say that the meter will attempt to overexpose fully saturated red by about half a stop.
That is just about a guarantee of a clipped red.
Please note that the screenshot of the red appears to be not full saturation but the developed raw to jpg shows clear clipping.
This experiment is fully repeatable.
If you wish to dispute the results then repeat the test and show that I am wrong.
I have to say I was surprised how clear the result was.

[savoche]

Interesting. Thanks for taking the time to do a systematic test!

I have to say I was surprised how clear the result was.

I don't really find your results surprising as they correspond well with what we all see in practice. I have long suspected that the meter sensor is "blind" to reds in some way, but your findings would provide a better explanation. And the workaround is simply to underexpose half a stop whenever shooting an image where reds dominate.

And blues, I suppose, although I haven't noticed the same tendency to blow the blues. It might be simply because I don't encounter pure blue subjects very often - or that the sensor meters blues better than it meters reds. Would be interesting to test some more.

Again, thanks for an interesting test!

[TonyW]

....So I was suspecting the meter was essentially seeing a greyscale of the colours....

The meter cannot see anything other than luminance values. Spot metering will be the most accurate/consistent method of reading luminance values. It may or may not be more/less sensitive to certain colours

Now have a look at the colour wheel once it is greyscaled – the reds and blues have ended up with a tone that is darker than the 18% grey that meters tend to calibrate to.

Meters are not calibrated to 18% they never have been since standards introduced. Average calibration point will be around -3EV from sensel saturation = 12.5%. Try it yourself meter something with texture and from the indicated exposure make several more increasing 1/3 stop each time until you are at least +3.67 EV from meter reading. If meter is 12.5% you will have hit saturation at +3EV, if meter calibration is 18% then you will only have +2.47EV to saturation.
Important to note you will not see accurate clipping response in any (AFAIK) raw editor due to the way images are rendered as covered in a previous post. Adobe add BLE and also add exposure, contrast and black level over their previous standards. Others may differ! You will need ot examine your file with Rawdigger or similar to get a proper indication of clipping

Could it be that the meter is metering red (and blue) this way and so giving a totally different exposure to them compared to how it would meter white.

The colour sensitivity of the meter may be less than even across all colours but it is the lack of red and blue sensels (1/2 that of green) that really make a difference and is one of the winning points with pixel shift systems.

With respect your conclusions appear to be founded on several misunderstandings:

1. That camera exposure meters calibrated to 18%. Generally this is not the case with 12% being more normal. The difference representing 0.5EV i.e. meter a so called 18% grey card and for a 'normal' subject you will need to increase the indicated exposure by +0.5EV or angle the card so that it is 45 degrees to the light - same effect.

2. 18% grey is actually the middle grey of exposure and should be close to RGB values of 128. There is no colour editing space in common use where 128 represents a mid point. Use L*a*b* to get a measure of mid tone set 50,0,0 and you will find that values change for instance I usually work in Prophoto with raw files and middle grey is about RGB =100, Adobe RGB is RGB=118 and sRGB close enough

3. Not familiar with your raw editor but there is no information about what it adds to raw to first render and image but one thing is certain it does not provide you with a real view of your raw data just its rendering and resulting histogram

4. You cannot rely on the EV controls in a raw converter to mimic what happens with exposure changes in camera. Your camera is an analogue capture device with an A/D converter turning photon count into digital numbers in a linear space. Your raw editor must apply a number of adjustments including, demosaicing, finding white and black points, finding a decent White balance, and importantly apply a gamma curve. When you apply exposure correction in your raw editor you are applying through a TRC (gamma curve) which will roll of highlights and shadows very differently from your original capture.

5. Clipping in JPEG? If you have not clipped in raw then your conversion to JPEG shows clipping then that is an error in your editing. Perhaps I have misunderstood the relevance of mentioning JPEG as the discussion really is raw capture

You may or may not be correct that your cameras meter is a little hot but I would expect that you will get similar if you care to shoot a bright blue subject.

In the other thread I asked if anyone was prepared to actually post a link (dropbox maybe) of red problem for us to look at the actual data - that offer is still open

[stevebrot]

I was playing around a bit with all this yesterday and determined (not surprisingly) that metering from a gray card was a huge preventative measure. If one wants to us the in-camera histogram (i.e. JPEG rendering) as a guide, custom white balance from a gray card reading is also quite helpful. Of course neither of those is a surprise. I would have tried in-camera spot metering too except that I have an aftermarket screen and spot meter is off the menu as a result.

In the other thread I asked if anyone was prepared to actually post a link (dropbox maybe) of red problem for us to look at the actual data - that offer is still open

Perhaps you might suggest the proper experimental conditions for a valid result? Or is it enough to simply post a DNG with full saturation of the reds?


Steve

[TonyW]

I was playing around a bit with all this yesterday and determined (not surprisingly) that metering from a gray card was a huge preventative measure. If one wants to us the in-camera histogram (i.e. JPEG rendering) as a guide, custom white balance from a gray card reading is also quite helpful. Of course neither of those is a surprise. I would have tried in-camera spot metering too except that I have an aftermarket screen and spot meter is off the menu as a result.

😄 I am not surprised that you are not surprised about grey card reading with modern DSLR’s 😉. They have their place perhaps under certain special circumstances and in ‘their day’ served as a useful central pivot point in a 10 stop analogue B&W world, with the caveat that 18% needed correction to 12.5% by +1/2 stop. Shooting reversal film I never found a grey card of particular use although I did use palm of hand to meter off and increase exposure accordingly to prevent blown highlights.

Perhaps you might suggest the proper experimental conditions for a valid result? Or is it enough to simply post a DNG with full saturation of the reds?

I would not want to be too prescriptive about ‘proper’experimental conditions but there are limits as I know that you know. I think inadvertently this post has actually proved a point I was trying to make ( labour !) 😩 here and in the other thread - I will wait for further reply rather than expand now.

A comment was made about no recovery from clipping a channel and this to a large degree is correct, but only if the channel truly clipped and that confirmation can only be had in a program that reads the values in the RGBG channels without applying correction.

A further suggestion that recovery not possible from a clipped channel. This may also be the case but is far from the rule in my experience if photo restorations.

To answer your question directly, a raw with a full saturation of reds would be interesting as would one showing definite clipping in any channel. No promises of miracles of course 😉😁

[GUB]

Tony W -- Here is the 7 test images for the test to establish clipping point. The 7th image was so clipped I didn't bother including it on the post. If you can't find a clipped red here then we are on different planets!
Have fun with Rawdigger.
Dropbox - reds - Simplify your life

[GUB]

3. Not familiar with your raw editor but there is no information about what it adds to raw to first render and image but one thing is certain it does not provide you with a real view of your raw data just its rendering and resulting histogram

Darktable. Base curve and Highlight reconstruction and White Balance were left enabled. Disabling Base Curve did not make much difference to the point at which the histo clipped but rather just shifted the histo around. Disabling highlight reconstruction just made everything clip - I suspect it required input from the Dynamic Range - highlight correction from the camera to set its values but I had that switched off.

5. Clipping in JPEG? If you have not clipped in raw then your conversion to JPEG shows clipping then that is an error in your editing. Perhaps I have misunderstood the relevance of mentioning JPEG as the discussion really is raw capture

No JPG editing here - all screenshots of raw editing laid out in a jpg image courtesy of Gimp.

You may or may not be correct that your cameras meter is a little hot but I would expect that you will get similar if you care to shoot a bright blue subject.

I am in no way suggesting that my meter is "hot"
What I have presented is evidence that in these conditions the meter can present 6 stops worth of different values for the same exposure. I expected it to be 2 to 3 stops as in film days but 6 stops !! -- I can guess the incidence exposure better than 6 stops.
I presume modern matrix metering can sense the tonal range in a mormal image and lay out the exposure more accurately. But in these black/white/red images there is no tonal range for the meter to work on. Using the spot meter will increase the likely hood of there being no tonal range.

[TonyW]

Tony W -- Here is the 7 test images for the test to establish clipping point. The 7th image was so clipped I didn't bother including it on the post. If you can't find a clipped red here then we are on different planets!
Have fun with Rawdigger.
Dropbox - reds - Simplify your life

GUB, thank you for the images.

Only had time for a cursory glance but you do seem to be on another planet there is no red clipping even in the lightest image it is well under clipping point examining in RawDigger and LR. In any digital capture there are a finite number of levels, for your K1 the numerical value for saturation is 16383 for the RGBG channels. The value for your lightest image for red channel in the mid 15000’s so not reached full well capacity. Obviously all other channels well below saturation.

There is a difference in WB and an obvious hue shift but not what you so far have described and all very easily corrected to match. Please send me a clipped one

I will check properly tomorrow

[GUB]

GUB, thank you for the images.

Only had time for a cursory glance but you do seem to be on another planet there is no red clipping even in the lightest image it is well under clipping point examining in RawDigger and LR. In any digital capture there are a finite number of levels, for your K1 the numerical value for saturation is 16383 for the RGBG channels. The value for your lightest image for red channel in the mid 15000’s so not reached full well capacity. Obviously all other channels well below saturation.

There is a difference in WB and an obvious hue shift but not what you so far have described and all very easily corrected to match. Please send me a clipped one

I will check properly tomorrow

Isn't the hue shift a bit of a giveaway that the greens and blues have continued amplification while the red is stuck at full saturation.?
Yes I can reshape the end of the histo with the Highlight reconstruction function but I suspect it is a random rearrangement of the full value red pixels.
Do you really think Pentax has kept a couple of extra stops of dynamic range up its sleeve in the image output without telling anyone?

[TonyW]

Isn't the hue shift a bit of a giveaway that the greens and blues have continued amplification while the red is stuck at full saturation.?
Yes I can reshape the end of the histo with the Highlight reconstruction function but I suspect it is a random rearrangement of the full value red pixels.
Do you really think Pentax has kept a couple of extra stops of dynamic range up its sleeve in the image output without telling anyone?

Well I am going to turn in now but suggest you check your testing methodology as it seems quite flawed from the very first post.

There is no red stuck at full saturation in your images and there is no sign of hitting the brick wall that is clipping. In short there is nothing that is outside of recovery

[stevebrot]

I would not want to be too prescriptive about ‘proper’experimental conditions but there are limits as I know that you know. I think inadvertently this post has actually proved a point I was trying to make ( labour !) 😩 here and in the other thread - I will wait for further reply rather than expand now.

A comment was made about no recovery from clipping a channel and this to a large degree is correct, but only if the channel truly clipped and that confirmation can only be had in a program that reads the values in the RGBG channels without applying correction.

...and that program might be?

It is fairly easy to sit back and say, "Fail" and again "Fail", and again "Fail". It would be nice to know the definition of a channel being "truly" clipped. What I can do is provide a short list of considerations at work at the sensor level that might temper the discussion:

• Photographic camera sensors are tuned to provide reasonably flexible raw spectral response from commonly encountered light sources (meaning voltages to the A/D converter, not pixel data)
• The above does not mean all light regimes are gracefully addressed or tonal gradation for all values of any color for that matter
• Nobody is suggesting that one will see prematurely clipped red channel from white areas of the subject in commonly encountered light regimes (e.g. daylight, cloudy, blue hour, golden hour, etc.)
• It is acknowledged that the appearance or histogram of a JPEG or TIFF created through normal means may bear processing artifact with or more channels clipped. This may be traced do data scaling and/or white balance adjust and/or the incantations of contrary sorcerers.
• It is easy enough to demonstrate a max channel value for a particular pixel, but not quite so easy to show that meaningful data were swallowed in its creation
• Images in bracketed exposure series of a subject having fine tonality (details) and high primary color saturation will show loss of data, but the source is difficult to assign to "true" clipping at capture
• Histograms using extracted PEF/DNG capture data from such a series might be helpful

I hope this helps.


Steve

[stevebrot]

for your K1 the numerical value for saturation is 16383 for the RGBG channels

That is the theoretical value. The assumption is that the chip supports the full 14 bits for all channels.


Steve

[stevebrot]

There is no red stuck at full saturation in your images and there is no sign of hitting the brick wall that is clipping. In short there is nothing that is outside of recovery

Does that mean that you are willing to attempt recovery?


Steve