“Expose to the Right” & relation to ISO Invariance

[digitaldog]

How did you get the 100 ISO and 800 ISO images to look the same?  Underexpose the 100, the increase "exposure" in post?

See below, already asked and answered.

[jrsforums]

See below, already asked and answered.

Where???

[digitaldog]

Where???

Reply #61 on: June 17, 2015, 10:33:13 AM »

[jrsforums]

Reply #61 on: June 17, 2015, 10:33:13 AM »

Found it

I'll post the facts, you Gentlemen can argue, discuss, come to conclusions.

I provided two captures.
Capture 1 was shot at ISO 100 with a f-stop and shutter recommendation from an Incident meter (Minolta Flash Meter III). It is what many would suggest is 'correct' exposure.
Capture 2 was shot at ISO 800 with the same settings as capture 1. Identical in every respect expect I changed the ISO setting from 100 to 800!
Capture 2 shows less noise than capture 1. This is a very magnified view of the image but it is clear that the capture with the higher ISO has less noise, better S/N.

I provided this example to dismiss two commonly heard urban legions. The first is higher ISO always produces more noise. That's clearly not the case here!
The other reason I posted was to dismiss the idea that ETTR must be conducted at base ISO. If that's true, explain the effect upon the noise at ISO 800.

Now you guys can argue, discuss and come to some conclusions in terms of semantics  about what Exposure means, what ETTR means, or the benefits expected from ETTR (I'll suggest the expected benefits are exactly what is seen in capture 2 and not capture 1 despite the external meter told me capture 1 was the 'correct exposure'.

You did not answer how, with 3 stops difference, you got the images to look as if they had the same exposure....what did you do in post?

[digitaldog]

Found it

Wrong post. 61, not 57.

[jrsforums]

Wrong post. 61, not 57.

What changes were made to the 800?

[bjanes]

need to spell raw converter parameters too...

That's right. I was referring to the sRGB image rendered by the camera into sRGB. ACR/LR are not good tools for evaluating the raw file levels because of the BaselineExposure offset, automatic highlight recovery (with PV2012), and tone curve. Even better, one should check the raw levels with RawDigger. The nominal exposure should give sensor saturation of ~18% for Canon, which does not allow 0.5 EV highlight headroom. Nikons give ~12% saturation, allowing 0.5 EV headroom for the highlights.

Bill

[bjanes]

when dealing with Adobe converters do use gutted .dcp profiles (linear curve, hidden expocorrections compensated, post exposure corrections LUTs removed) + Process 2010 + brightness 50, contrast 25, curve : Medium Contrast, output to the baddest gamut-wise colorspace... something like this... at least in ACR.

In my experiments with evaluating the raw file with ACR, I get best results with PV2010, linear tone curve, and the sliders on the basic panel all set to zero. I use the Adobe Standard camera profile. These are the settings recommended by the ICC for obtaining scene referred data.

Bill

[AlterEgo]

I use the Adobe Standard camera profile

and I don't (for the reasons outlined above about expocorrection, etc) - hence I use different parameters...

These are the settings recommended by the ICC for obtaining scene referred data.

recommended when CS3 was the current version, that will be ~2007 ? before Adobe introduced what we have now in their DCP profiles now I think...

[bjanes]

and I don't (for the reasons outlined above about expocorrection, etc) - hence I use different parameters...

recommended when CS3 was the current version, that will be ~2007 ? before Adobe introduced what we have now in their DCP profiles now I think...

I forgot to add that with my D800e, one must use a negative exposure correction of -0.35 EV. I should probably use a profile that definitely lacks a twist. Since I am dealing with a Stouffer step wedge and other neutral targets, color is not that important and my results with ACR 9.0 and PV2010 with a linear tone curve as outlined in the ICC post and with Adobe Standard correlate quite well with DCRaw conversions with a linear tone curve and with readings from RawDigger. What ACR profile do you recommend? If you render into ProPhotoRGB you have to deal with a gamma encoded space and this complicates checking for linearity. However, a log-log plot will allow a linear plot with gamma encoded data. A nice feature of ACR 9.x is that it permits rendering directly into Linear RIMM.

As to your reference to 2007 and CS3, my testing with current software shows that one can still get excellent results. PV2010 with the default parameters that you quote will not give a linear response. Have you plotted your data with these parameters? I would be interested in seeing a scatter plot showing exposure and rendered pixel values over a wide range, say from near clipping of highlights to -6 or -7 EV. Shown below are scatter plots showing ACR renderings into ProPhotoRGB using Adobe Standard. PV2010 with the defualt tone curve using medium contrast in the point curve and brightness and contrast values of 50 and 25 does not produce a linear curve. PV2010 with the point curve set to linear and the sliders on the basic panel does work well.

Regards,

Bill

[Jack Hogan]

Let's try this again. All I want is:
A. One good sentence or paragraph to describe what happened with the ISO 800 image that resulted in less noise.

Cameras that are not ISOless/invariant are not able to transfer to the raw data the full dynamic range collected at the photosites during the given exposure in one go, because the electronics have a smaller DR than the sensor (say 12 vs 14 stops).





By changing ISO with the same exposure you are choosing what portion of the scene information available at the photosites to store in the raw data. If you raise ISO, you are choosing to transfer shadow information to the raw data at the expense of highlights:

B. Agreement on what to call this 'technique' other than ETTR since it's presumably not a part of exposure.

Given the electronics' more limited DR, raising ISO at a given Exposure favors tones in the shadows for storage in the raw data at the expense of highlight headroom.  Call it 'Sensor Tone Range Optimization' or similar.

C. Agreement on whether what I see by upping the ISO is beneficial (it appears to me that it is).

Sure it is, it is part of tone selection and optimization by the photographer with an ISOful camera in the field.

On the other hand an ISOless camera does not require such worries: after maximizing exposure just stick with base ISO because it is able to store most (all) information available at the photosites into the raw data in one go.





Raising ISO in the ISOless case simply throws away highlight headroom with no corresponding benefit in the shadows.  Images from articles here and  here:

Jack

PS How can I make in-line images smaller?

[bernie west]

All I want is:
A. One good sentence or paragraph to describe what happened with the ISO 800 image that resulted in less noise.

Here's what I wrote earlier while answering my own similar question (in relation to the non-ISOless 5DIII).  Others confirmed this is right:

A possible answer to my own question... Does this mean that the majority of electronic noise is introduced into the image data in the ADC? [YES]  That would mean that boosting the ISO in non-ISOless cameras would boost the signal in isolation from the majority of the electronic noise.

[sandymc]

Something to be aware of is that when going to in-camera JPEG, many cameras apply noise reduction. That noise reductions can vary by ISO (or by analysis of the image), so higher ISO, more noise reduction. I don't know whether a 5DIII does that or not - I don't pay much attention is in-camera JPEGs in what I do. But that would be my first guess as an explanation for the difference in posted images.

Sandy

[bernie west]

He developed it from raw.

[AlterEgo]

I forgot to add that with my D800e, one must use a negative exposure correction of -0.35 EV. I should probably use a profile that definitely lacks a twist. Since I am dealing with a Stouffer step wedge and other neutral targets, color is not that important and my results with ACR 9.0 and PV2010 with a linear tone curve as outlined in the ICC post and with Adobe Standard correlate quite well with DCRaw conversions with a linear tone curve and with readings from RawDigger. What ACR profile do you recommend? If you render into ProPhotoRGB you have to deal with a gamma encoded space and this complicates checking for linearity. However, a log-log plot will allow a linear plot with gamma encoded data. A nice feature of ACR 9.x is that it permits rendering directly into Linear RIMM.

As to your reference to 2007 and CS3, my testing with current software shows that one can still get excellent results. PV2010 with the default parameters that you quote will not give a linear response. Have you plotted your data with these parameters? I would be interested in seeing a scatter plot showing exposure and rendered pixel values over a wide range, say from near clipping of highlights to -6 or -7 EV. Shown below are scatter plots showing ACR renderings into ProPhotoRGB using Adobe Standard. PV2010 with the defualt tone curve using medium contrast in the point curve and brightness and contrast values of 50 and 25 does not produce a linear curve. PV2010 with the point curve set to linear and the sliders on the basic panel does work well.

Regards,

Bill

certainly you need to use the output colorspace that suits you - if you want to use (which you can) the widest gamut and linear gamma RGB colorspace for your purposes as you noted ACR will allow to select one - just supply the icc/icm - you know that... if your task is to imitate rawdigger more precisely by all means use whatever it takes - but do suggest to remove everything except the matrix parts (to make sure that there are no twists and brightness corrections through LUTs), insert a specific linear tone curve in the profile just in case (in case the profile do not have the curve ACR will use its/yours own settings, but I rather make sure that the curve is in the profile too), and make sure to use BaseLineExposureOffset compensating BaseLineExposure - you know all that too...

my use is different - I mostly use the settings to see how custom DCP profiles fit the measured targets, for example



this shows a decent fit to a measured target with L* from measurements = 96.89, 81.26, 66.68, 50.64, 35.79, 20.50 / readouts in ACR with custom made profile and settings that I used are = 97, 81, 67, 51, 36, 20.

and RawDigger shows that clipping in raw, green channel, is around 1/6 (or rather closer to even 1/3EV if we use the center/top of the "peak", instead of the very right edge of it in histogram for a selected area on a white patch till clipping for Sony A7 that I think is 15860 or so) EV from what we have in the whitest patch - which corresponds good enough with Adobe's visual display of their "histogramm" here

PS: damn Adobe for not supplying Lab readouts with 0.1 precision - just like Iridient did overkill it with 0.01 precision, Adobe still managed to spoil the goods with just integer values for Lab readouts...

[sandymc]

He developed it from raw.

I didn't see any raws posted.......

[digitaldog]

Cameras that are not ISOless/invariant are not able to transfer to the raw data the full dynamic range collected at the photosites during the given exposure in one go, because the electronics have a smaller DR than the sensor (say 12 vs 14 stops).

Thanks Jack, let me look this over, digest it etc.

PS How can I make in-line images smaller?

I think the size is great, easy for us older farts to see.

[Guillermo Luijk]

You cannot check ISO noise by changing ISO, then adjusting in post.  You must reduce the exposure one stop for each stop increase in ISO.

Completely incorrect approach. To evaluate the effect in SNR of a variable (ISO), that variable must be isolated from any other (exposure in this case). This is a basic scientific principle.

Changing ISO and exposure at the same time (the usual procedure when taking pictures) leads to confusing and misleading results, e.g. the "more noise the higher the ISO" legend, which is wrong. Visible noise increases when exposure is reduced, not when ISO is pushed.

In some sensors (Canon type), pushing ISO increases SNR. In some others (Sony type, the so called isoless sensors), pushing ISO has little effect in improving SNR.

Whether ETTR can be achieved by pushing ISO (when available exposure doesn't suffice) or not, is just a semantic discusion. The true is that for a given exposure, some sensors improve their SNR (the only goal of ETTR) the higher the ISO:

Canon 350D at constant exposure:



Regards

[digitaldog]

Changing ISO and exposure at the same time (the usual procedure when taking pictures) leads to confusing and misleading results, e.g. the "more noise the higher the ISO" legend, which is wrong. Visible noise increases when exposure is reduced, not when ISO is pushed.

Excellent, thanks. I was hoping you'd find your way into the discussion. Your illustration just provided was the original example I recall seeing that lead to my attempt testing ISO on my Canon.

Whether ETTR can be achieved by pushing ISO (when available exposure doesn't suffice) or not, is just a semantic discusion.

Agreed although it would be great if we could come to some common agreement about the language. It might not be possible. 

[Bart_van_der_Wolf]

In some sensors (Canon type), pushing ISO increases SNR. In some others (Sony type, the so called isoless sensors), pushing ISO has little effect in improving SNR.

Whether ETTR can be achieved by pushing ISO (when available exposure doesn't suffice) or not, is just a semantic discusion. The true is that for a given exposure, some sensors improve their SNR (the only goal of ETTR) the higher the ISO:

Canon 350D at constant exposure:

And assuming the ISO 100 was correctly exposed, now the (cropped out) highlights are totally clipped. By underexposing the ISO 100 shot 16x (so we create 4 stops headroom for the subsequent amplification), we then and only then can improve the S/N ratio on some cameras without clipping highlights, like the Canons, by increasing the ISO setting 16x. However, bracketing exposures creates much better image quality for such stationary subjects, and just a bit more work in postprocessing.

Cheers,
Bart