[TDvN57]

I few days or weeks ago (I cant remeber when exactly) I read here on PF about a discussion to shoot only at 100 ISO and in low light to rather under expose and then apply the needed gain in post processing. The rationale was that the post processing gain can be done with better results than adding the gain in the camera by adjusting the ISO upwards.

I did two test shots very quickly, pic #1 at 100 ISO at -6 stops under exposed and pic #2 at correct exposure with ISO 6400. At a glance it seems that the colour depth is much better with the exposure gain applied in ACR. The hisograms from ACR is placed on top of the pictures.

Any comments? Has anybody tried this as poart of a workflow and what was your experience from it?

645z + P67 M* 400mm

Attached Images

View Picture EXIF

File Date

 

View Picture EXIF

File Date

 

[AfterPentax Mark II]

I few days or weeks ago (I cant remeber when exactly) I read here on PF about a discussion to shoot only at 100 ISO and in low light to rather under expose and then apply the needed gain in post processing. The rationale was that the post processing gain can be done with better results than adding the gain in the camera by adjusting the ISO upwards.

I did two test shots very quickly, pic #1 at 100 ISO at -6 stops under exposed and pic #2 at correct exposure with ISO 6400. At a glance it seems that the colour depth is much better with the exposure gain applied in ACR. The hisograms from ACR is placed on top of the pictures.

Any comments? Has anybody tried this as poart of a workflow and what was your experience from it?

645z + P67 M* 400mm

What I notice is that picture #1 is sharper (or the second one is softer) and therefor colours seem more brilliant. Words in colorchecker are sharp in the first and not in the second. I do not know if at 6400 iso the whole picture is negatively influenced by that iso.

[TDvN57]

What I notice is that picture #1 is sharper (or the second one is softer) and therefor colours seem more brilliant. Words in colorchecker are sharp in the first and not in the second. I do not know if at 6400 iso the whole picture is negatively influenced by that iso.

I unfortunately I bumped the camera when I took pic #2 @ 6400 ISO so the focus is off. I think the histogram is the most reliable indicator that shows the difference between the two pictures.

[biz-engineer]

Not sure about 645z. If I remember correctly, the camera applies some noise reduction to raw data and perhaps equalize r.g.b channels at higher ISO exposures, and blown high-lights can't be recovered. Shooting underexposed at base ISO let more noise, more detail, potentially some color cast, but leaves more room for highlights. At low ISO underexposed , all you see is a dark image when you chimp (review image just taken).

[Bob 256]

The invariant ISO claim for post processing equivalency applies to RAW images, so if you're using JPEG images, there could be some issues involved depending on how the camera processes the image to get the JPEG file.

[othar]

The invariant ISO claim for post processing equivalency applies to RAW images, so if you're using JPEG images, there could be some issues involved depending on how the camera processes the image to get the JPEG file.

I am pretty sure it was tested with RAW files, 6 stops of light would have been too much for jpgs.

[Bob 256]

I am pretty sure it was tested with RAW files, 6 stops of light would have been too much for jpgs.

Now going back and re-reading the post, I don't know where I got the idea that JPEG was used - my bad.

[ProfessorBuzz]

Tough question.

Dialing up the "gain" in post is likely just multiplying up the pixel values, so that will dial up the noise as well as signal; and I think you can see the histogram compression going on.
Increasing ISO (which increases the on-chip gain selected in the CMOS APS sensor) will bring up the low signal levels.
Say you have low in-camera gain, so the darkest areas barely register. No amount of post gain will bring out detail that isn't detected. (Put another way: Multiply 0 x 10000% post gain, you still get 0).

Dialing up the sensor's gain may get you a few more ADU counts in these darker areas; then you can tease out more detail.

In science cameras, we'll do "stacking" which co-adds repeated images; signal will rise in proportion to sqrt(noise) so it is more effective.

[TDvN57]

Tough question.

Dialing up the "gain" in post is likely just multiplying up the pixel values, so that will dial up the noise as well as signal; and I think you can see the histogram compression going on.
Increasing ISO (which increases the on-chip gain selected in the CMOS APS sensor) will bring up the low signal levels.
Say you have low in-camera gain, so the darkest areas barely register. No amount of post gain will bring out detail that isn't detected. (Put another way: Multiply 0 x 10000% post gain, you still get 0).

Dialing up the sensor's gain may get you a few more ADU counts in these darker areas; then you can tease out more detail.

In science cameras, we'll do "stacking" which co-adds repeated images; signal will rise in proportion to sqrt(noise) so it is more effective.

I see your point, although the signal value registered in the individual pixel would still be the same. Whether the gain is applied in camera during the analog to digital conversion, or if you apply the gain in post, neither of these two scenarios change the number of photons received which resulted in the signal voltage registered in the pixel. Stacking the signal is obviously different, since you are adding photons which in concept it will be the same as a longer exposure. (I know the practicality of stacking versus long exposure determines which is the best solution, but I am trying to just consider the process from sensor onwards).

What I am trying to determine is for the same signal strength captured on sensor, is there an advantage to let the camera add the gain by simulating a higher ISO value, or is the post processing software like ACR better equiped to add the gain? My gut feel is that in camera is always better, but that may not be true anymore. Post software has the potential to have an advantage in that it has a lot more processing power available (our computers for example) and new concepts can be applied and rolled out easier than with firmware updates to a camera, potentially resulting in a better image quality.

I suppose the nutshell question would be if the potential of the post software is real or even possible. Perhaps the answer is also camera dependant.

Maybe someone has already gone down this path with answers, which will be very helpful if shared.

I'll take a few more shots with the 645z and share the results.

[kaseki]

If detectors had no noise, there would still be a difference between strong illuminance images and weak illuminance images with gain. This is due to the "signal" (photodetector charge from the incoming light) having noise itself due to its quantum nature. The RMS noise is the square root of the signal charge count (photo electrons in the "bucket). The specific signal noise in a given pixel is a value within a Gaussian distribution (Poisson distribution for low average levels). Hence, a weak uniform image scaled up will show noise that won't be as visible in an initially stronger image. S/N = square root of S. Larger S, better S/N.

[TDvN57]

I took a few more pictures to compare the gain in camera vs in post. While at it I decided to add the in-camera noise reduction as a comparisson as well. I used the 645z + FA 120mm macro and the X-Rite colour checker as a target.

Each picture below has four pictures.

Top two pictures taken with Noise Reduction set to Auto in camera
Top Left: Aperture f/16, ISO & Shutter speed adjusted to set exposure.
Top Right: Aperture f/16, ISO fixed at 100, Shutter speed adjusted to set exposure. Gain to match ISO exposure added in ACR.

Bottom two pictures taken with Noise Reduction set to Off in camera
Bottom Left: Apertue f/16, ISO & Shutter speed adjusted to set exposure
Bottom Right: Aperture f/16, ISO fixed at 100, Shutter speed adjusted to set exposure. Gain to match ISO exposure added in ACR.

Unfortunately the light changed between taking the top row of pictures versus the bottom row. Hence the bottom row had to be exposed 1/2 stop longer which resulted in the bottom row having slightly more contrast. It can also be seen in the histogram in the first picture.

I suggest the interpretation of the pictures and histograms to be between left (in camera gain) and right (post gain in ACR). These were taken under similar light conditions.

The pictures were taken from ISO 100 at correctly exposed. Each picture would we exposed less by one stop. The left picture panel the ISO and shutter speed is adjusted by one stop. The right panel the ISO is left at 100 ISO and the shutter speed is adjusted faster by one stop, being under exposed.

The full range from ISO 100 (0 under exposed) to ISO 205k (11 stops under exposed).

There does not seem to be much obvious difference between gain in camera vs gain in post up to 8 stops under exposed (25k ISO).

My conclusion is that the in-camera processing is far superior and does a great job. It is further improved to switch on the in-camera noise reduction, which really kicks in from ISO 25k onwards.

Any other info to learn from this?

[kaseki]

Bravo! This ability to work at low light levels was a significant factor in my decision to purchase the 645Z. Because the charge on the electron is 1.6E-19 coulombs, even very weak illumination still provides a large enough number of photoelectrons that the S/N = root S is decently high.

[TDvN57]

Bravo! This ability to work at low light levels was a significant factor in my decision to purchase the 645Z. Because the charge on the electron is 1.6E-19 coulombs, even very weak illumination still provides a large enough number of photoelectrons that the S/N = root S is decently high.

What surprised me the most was the first picture that prompted me to start the discussion and that I couldn't repeat the results. When I get time I'll do a few more tests with a natural view and wider colour and luminance range and see if there is a bandit to consider post process gain vs in-camera gain.

But yes, the 645z can perform well in low light with high iso, but it takes learning. So far it seems not to under expose is one requirement, I.e. bias +/- set to 0, metering set to matrix (maybe, not sure yet). I'm still learning.

[TDvN57]

I did a few more tests and found that if I limit the at ISO 3200 in low light conditions, then adjusting the "gain" in post production I got very similar results compared to with out of camera results, up to 52k. Beyond that the picture falls apart much worse than the out of camera results.

So to revise my previous conclusion: underexposing a low light picture does not make it worse, it does not make it better either. Beyong 52k ISO the in camera result is better. Not that I would take a serious picture at ISO 52k, unless it happens to be a picture of bigfoot.