Too little dynamic range and noisy shadows

Reflector

Mu-43 Hall of Famer
Joined
Aug 31, 2013
Messages
2,283
Hope the title of the topic caught everyone's attention since that is what I hear about Micro Four Thirds with all the recent releases from Canon, Nikon and Sony...

Excuse this heavily overprocessed shot that didn't make it outside of my unprocessed collection of shots that I consider mediocre, I'm making a point here since there has been this constant noise from the latest releases that seem to beat a dead horse...

Also little bit of incompleteness on the blend but the 100% is where the real illustration is at.

I can't help anyone that is trying to take photos of BIF or other very, very dynamic subjects and wants a to wipe out noise easily. Get a D5/1DXII/A9/A7SII for that and the fastest f/1.8-2.8 primes/zooms possible.

PA023386N.jpg
Subscribe to see EXIF info for this image (if available)

A 100% @ 45mp crop below:
PA023386NC.jpg
Subscribe to see EXIF info for this image (if available)


What it was made out of:
Two Shots.jpg
Subscribe to see EXIF info for this image (if available)

The camera does not control you and hopefully the new camera releases do not control you either. If either or both do then seek help immediately.
You control the camera, you're in charge of utilizing the camera in an optimal manner to try to push yourself to the limits.

Made out of two HR exposures, treated differently than I have handled previous multiple exposure median blending. This is closer to what I found myself doing for two shot HDRs. I could actually stack the second shot a third time but the gains are too little given it was intentionally exposed dark (and accidentally at ISO800... whoops) and any if little motion needs to be masked away.



It is possible to "outshoot" larger sensors significantly results by using bracketing in addition to normal stacking and in a manner more than the typical median stack.

Based on the same principles here using bracketing and IBIS, a 3 or 5 shot bracket at 60fps would effectively allow for handheld "HDR" while also reducing noise. Hopefully this controversial topic title and shot was enough to grab some attention. The hardware in the cameras is more than capable

Computational imaging and what will lead to an almost "noise free" shot in the shadows so long as you have sufficient shutter speed.

I've tested this out by doing test shots indoors using the bright exterior and normally illuminated interior, but I have yet to go to a particularly notable/scenic location to actually take shots for this. When I have time I will put a practical second example up in the next post which I'll reserve.

As a bonus shooting in RAW isn't absolutely critical for doing this but it certainly helps.


Shooting:
  1. Set drive to continuous high with the electronic shutter (You don't need a G9/E-M1II, sufficiently fast drives makes life easier)
  2. Enable bracketing (3 or 5 exposures, 1EV steps)
  3. Using the histogram/blinkies (Set the highlight to 254 or a lower number)*, dial down the exposure compensation to see how much underexposure is needed to prevent any blown highlights but leave it still intentionally exposed very bright. Note the number (Say, -1.3EV or -0.3EV...) and remember your shutter speed so you don't blur the overexposed frames.
  4. Make sure the most underexposed shot in the bracket is set to that value. Let the camera do the work and take all your shots in rapid succession.
*At "ISO Low" on the E-M1II, you have close to no headroom as the camera is doing a 1.7EV ETTR already. To figure out how much exposure latitude you have, you can leave the highlight blinkies at 250-255, the moment you see anything blink as a massive patch then you have a very high chance of unrecoverable/totally clipped the highlights. Pull the exposure back until you see a good chunk of the blinkies turn into a few. You have around -0.3 to -0.5EV to work with, so exposing 0.3 to 0.7 lower in ISO low is acceptable. Ignore specular highlights/bright light sources as those are always going to blow, just handle the roll off region around them instead.

At any other ISO(200-12800 in practice): The converse is true, you have roughly 1.3EV of headspace typically for most scenes. I shot my E-M5 this way and I often shoot the E-M1II this way.

Major adjustments in post processing:
  1. Go look at your multiple exposures, take your overexposed shots and pull the exposure back to the "normal" exposure. You should end up with 2 or 4 shots that look identical to each other aside from the progressively more burned out highlights
  2. Take the underexposed image with the intact highlights, make your adjustments ONLY for the highlights, darkening to make it look normal to your eyes. Ignore the rest of the image for how underexposed it is as you'll effectively discard those parts.
  3. Load all the images into layers in your favorite image processing program.

Stacking:
  1. Auto or manually align the images. IBIS should mean the actual amount of alignment is within 1 pixel at sufficiently high shutter speeds... In practice I know that down to 1/2s this is within reason still.
  2. Take the underexposed shot with the intact highlights, set it as the very top layer.
  3. The other shots, you want to set their blending so they are treated as a stack. The first layer should be 100%, the second 50% and then the third should be 33%, fourth 25%.
  4. If you feel like experimenting, you can intentionally make this a non median blend by biasing the more overexposed shots

Recovering highlights:
  1. Set the blending mode on the underexposed shot with intact highlights to darken or equivalent for non Photoshop users. The painful alternative is to manually mask in the shot, but this is the quick way.
  2. Enable masking, invert if needed to ensure that the underexposed disappears
  3. Brush, opacity/flow to less than 100% (tweak to your comfort levels) and then use the brush to brush down highlights. These could be the bright windows within a dimly lit room or anything else.

Quicker and faster than a HDR since you're just median blending the darker parts of the shot. The highlights are ETTRed for maximum data so those won't be noisy in practice. Motion artifacts can be ignored and if you feel like putting a little more time in, you can use the other shot(s) for static scenes by just lazily masking any areas of motion out.

2 shots yields:
Highlights are recovered
Motion artifacts can be avoided at the cost of losing median stacking advantages.

2.5 shots yields:
Highlights are recovered
Motion artifacts can be avoided at the cost of time for masking out the non overexposed shot (if possible on the moving subject)

3 shots yields:
Highlights are recovered + 1 normal shot + 1 shot overexposed
=
You're putting in the effective exposure data of 2.5x exposures. 1 exposure normal, 1 pushed (ETTR)

5 shots yields:
Highlights are recovered + 1 normal shot + 3 shots overexposed with significant data in shadows
=
You've got a ridiculous amount of exposure data, as you're exposing at 0+1+2+3, the most ETTRed shot (+3EV) will obliterate a ridiculous amount of noise in addition to the median or weighted blending. Metaphorically speaking you're shooting the equivalent of 15x* the exposure duration the normal + 3 overexposed shots. So an equivalent to "ISO <25" out of four ISO 200 shots or if one wants to think of it that way.

*Assuming the 0EV shot = 1s for calculation purposes, exposures respectively: 1s+2s+4s+8s = 15s effective exposure duration, assuming the underexposed shot is discarded.

Exposure wise, 5 shots fired off with a 60fps bracket has significantly more exposure (and thus data) than a 135 format camera taking a single exposure at the same exposure assuming that shot is exposed to avoid burning the highlights. 4 shots at the same exposure yields the equivalent amount of light to a 135 format sensor, but 5 shots from bracketing will yield even more data (and of the "signal" kind that reduces shot noise or the inherent behavior of light) in the shadow regions than a single exposure by far. The comparable amount goes almost to 4x exposures on the 135 format camera in the regions not blown out.

Almost 4x exposures worth is another way to say "almost a single exposure from 645 medium format."


More data is good. This is not limited to Micro Four Thirds but 60fps and incredibly effective IBIS sure makes life a lot easier for handheld shots.



Dealing with panning for dynamic subjects/dynamic backgrounds is doable, you will lose FoV via resolution due to cropping depending on the shots selected and stacked as a bit of a penalty. Otherwise the tradeoff will be made for noise at the edges where the stack does not align across several shots. Low motion subjects tend to the easiest to deal with, you want to try to maximize exposure on them without burning the highlights.

For anyone with a focal reducer and ridiculously fast and heavy lenses or have fast native lenses:
You've got a significant advantage as you will have higher shutter speeds.
 
Last edited:

Reflector

Mu-43 Hall of Famer
Joined
Aug 31, 2013
Messages
2,283
Not the promised practical example, but to give everyone an idea as to dealing with movement and what it looks like. Consider the below as having no worth in regards to photographic merit in subject/composition, it is merely for illustrative purposes with a moving object and was done quickly.

The two extreme scenarios, metering for the (impossible) highlights of the light and metering for the rest of the scene with ETTR:
1M283953DN.jpg
Subscribe to see EXIF info for this image (if available)


And no, the microfiber isn't blown out. The jpeg of it is, but the microfiber certainly isn't blown.

1M283953N.jpg
Subscribe to see EXIF info for this image (if available)

But when you're pushing to such extreme ranges, you're going to start seeing other problems like ghosting. In those cases making out the more ETTRed areas of trouble will take care of it.

What about the noise?
Ignore the exif data below, playing with layers does not make for accurate exif for 100% crops.
In order of the 5 shot stack (with not totally equalized settings), ETTR (maximized exposure) and ETTL (highlights preserved):
1M283953C.jpg
Subscribe to see EXIF info for this image (if available)
1M283953CETTR.jpg
Subscribe to see EXIF info for this image (if available)
1M283953CETTL.jpg
Subscribe to see EXIF info for this image (if available)


The most ETTRed file is still noticeably more noisy in the shadows but going up against a stacked 3 additional exposures will make for these kinds of comparisons.
 
Last edited:

Andrewmap

Mu-43 All-Pro
Joined
Jun 5, 2018
Messages
1,143
Location
Derby, United Kingdom
Real Name
Martin
Excuse this heavily overprocessed shot that didn't make it outside of my unprocessed collection of shots that I consider mediocre, I'm making a point here since there has been this constant noise from the latest releases that seems to beat a dead horse.

I can't help anyone that is trying to take photos of BIF or other very, very dynamic subjects and wants a to wipe out noise easily. Get a D5/1DXII/A9/A7SII for that and the fastest f/1.8-2.8 primes/zooms possible.

View attachment 680529
A 100% @ 45mp crop below:
View attachment 680530

What it was made out of:
View attachment 680528
The camera does not control you and hopefully the new camera releases do not control you either. If either or both do then seek help immediately.
You control the camera, you're in charge of utilizing the camera in an optimal manner to try to push yourself to the limits.

Made out of two HR exposures, treated differently than I have handled previous multiple exposure median blending. This is closer to what I found myself doing for two shot HDRs. I could actually stack the second shot a third time but the gains are too little given it was intentionally exposed dark (and accidentally at ISO800... whoops) and any if little motion needs to be masked away.



It is possible to "outshoot" larger sensors significantly results by using bracketing in addition to normal stacking and in a manner more than the typical median stack.

Based on the same principles here using bracketing and IBIS, a 3 or 5 shot bracket at 60fps would effectively allow for handheld "HDR" while also reducing noise. Hopefully this controversial topic title and shot was enough to grab some attention. The hardware in the cameras is more than capable

Computational imaging and what will lead to an almost "noise free" shot in the shadows so long as you have sufficient shutter speed.

I've tested this out by doing test shots indoors using the bright exterior and normally illuminated interior, but I have yet to go to a particularly notable/scenic location to actually take shots for this. When I have time I will put a practical second example up in the next post which I'll reserve.

As a bonus shooting in RAW isn't absolutely critical for doing this but it certainly helps.


Shooting:
  1. Set drive to continuous high with the electronic shutter (You don't need a G9/E-M1II, sufficiently fast drives makes life easier)
  2. Enable bracketing (3 or 5 exposures, 1EV steps)
  3. Using the histogram/blinkies (Set the highlight to 254 or a lower number), dial down the exposure compensation to see how much underexposure is needed to prevent any blown highlights but leave it still intentionally exposed very bright. Note the number (Say, -1.3EV or -0.3EV...) and remember your shutter speed so you don't blur the overexposed frames.
  4. Make sure the most underexposed shot in the bracket is set to that value. Let the camera do the work and take all your shots in rapid succession.

Major adjustments in post processing:
  1. Go look at your multiple exposures, take your overexposed shots and pull the exposure back to the "normal" exposure. You should end up with 2 or 4 shots that look identical to each other aside from the progressively more burned out highlights
  2. Take the underexposed image with the intact highlights, make your adjustments ONLY for the highlights, darkening to make it look normal to your eyes. Ignore the rest of the image for how underexposed it is as you'll effectively discard those parts.
  3. Load all the images into layers in your favorite image processing program.

Stacking:
  1. Auto or manually align the images. IBIS should mean the actual amount of alignment is within 1 pixel at sufficiently high shutter speeds... In practice I know that down to 1/2s this is within reason still.
  2. Take the underexposed shot with the intact highlights, set it as the very top layer.
  3. The other shots, you want to set their blending so they are treated as a stack. The first layer should be 100%, the second 50% and then the third should be 33%, fourth 25%.
  4. If you feel like experimenting, you can intentionally make this a non median blend by biasing the more overexposed shots

Recovering highlights:
  1. Set the blending mode on the underexposed shot with intact highlights to darken or equivalent for non Photoshop users. The painful alternative is to manually mask in the shot, but this is the quick way.
  2. Enable masking, invert if needed to ensure that the underexposed disappears
  3. Brush, opacity/flow to less than 100% (tweak to your comfort levels) and then use the brush to brush down highlights. These could be the bright windows within a dimly lit room or anything else.

Quicker and faster than a HDR since you're just median blending the darker parts of the shot. The highlights are ETTRed for maximum data so those won't be noisy in practice. Motion artifacts can be ignored and if you feel like putting a little more time in, you can use the other shot(s) for static scenes by just lazily masking any areas of motion out.

2 shots yields:
Highlights are recovered
Motion artifacts can be avoided at the cost of losing median stacking advantages.

2.5 shots yields:
Highlights are recovered
Motion artifacts can be avoided at the cost of time for masking out the non overexposed shot (if possible on the moving subject)

3 shots yields:
Highlights are recovered + 1 normal shot + 1 shot overexposed
=
You're putting in the effective exposure data of 2.5x exposures. 1 exposure normal, 1 pushed (ETTR)

5 shots yields:
Highlights are recovered + 1 normal shot + 3 shots overexposed with significant data in shadows
=
You've got a ridiculous amount of exposure data, as you're exposing at 0+1+2+3, the most ETTRed shot (+3EV) will obliterate a ridiculous amount of noise in addition to the median or weighted blending. Metaphorically speaking you're shooting the equivalent of 15x* the exposure duration the normal + 3 overexposed shots. So an equivalent to "ISO <25" out of four ISO 200 shots or if one wants to think of it that way.

*Assuming the 0EV shot = 1s for calculation purposes, exposures respectively: 1s+2s+4s+8s = 15s effective exposure duration, assuming the underexposed shot is discarded.

Exposure wise, 5 shots fired off with a 60fps bracket has significantly more exposure (and thus data) than a 135 format camera taking a single exposure at the same exposure assuming that shot is exposed to avoid burning the highlights. 4 shots at the same exposure yields the equivalent amount of light to a 135 format sensor, but 5 shots from bracketing will yield even more data (and of the "signal" kind that reduces shot noise or the inherent behavior of light) in the shadow regions than a single exposure by far. The comparable amount goes almost to 4x exposures on the 135 format camera in the regions not blown out.

Almost 4x exposures worth is another way to say "almost a single exposure from 645 medium format."


More data is good. This is not limited to Micro Four Thirds but 60fps and incredibly effective IBIS sure makes life a lot easier for handheld shots.



Dealing with panning for dynamic subjects/dynamic backgrounds is doable, you will lose FoV via resolution due to cropping depending on the shots selected and stacked as a bit of a penalty. Otherwise the tradeoff will be made for noise at the edges where the stack does not align across several shots. Low motion subjects tend to the easiest to deal with, you want to try to maximize exposure on them without burning the highlights.

For anyone with a focal reducer and ridiculously fast and heavy lenses or have fast native lenses:
You've got a significant advantage as you will have higher shutter speeds.

Reflector, thanks for a great tutorial - hope you don't mind but I have printed it off to read properly when I get home tonight.:clapping:
 

Machi

Mu-43 Top Veteran
Joined
May 23, 2015
Messages
798
I'm doing something similar in the FIJI. I found that simple summation of the source images looks often more natural than HDR.
(The image is from this thread)
mornn-jpg.jpg
Subscribe to see EXIF info for this image (if available)
 

Reflector

Mu-43 Hall of Famer
Joined
Aug 31, 2013
Messages
2,283
Still not a too practical example, but it demonstrates how much dynamic range you can pull out and how little noise is left afterwards.
Yes, it is very, very, very overprocessed and probably doesn't demonstrate the best handling of the highlights in the tree (I could expose by -0.3EV less)
Certainly not a "keeper" shot or "good" shot but it illustrates how much range you'll get out a bracketed stack without going through HDR. Adjustments can be made and tweaked to your own preferences. The biggest takeaway is you'll have an insane amount of latitude to play with afterwards.

Also some jpeg posterization might be occurring which I don't see in the 100%...


1M294078N.jpg
Subscribe to see EXIF info for this image (if available)


Below are the four shots out of the five bracketed shots for the sequence, the most overexposed shot is used for the movement compensation shot and I mask away slight movement with it. I was a bit messy and didn't mask down all the foliage so you can see the movement artifacts. When in doubt just overlay the ETTR shot, mask and brush liberally in the areas you think motion is at in any non-blown region.

Exposure speeds are, accordingly for the bracket: 1/200, 1/100, 1/50, 1/25 and 1/13. All shot in a single handheld exposure while standing, just stay steady and quickly review for motion blur.
1M294078 Sequence.jpg
Subscribe to see EXIF info for this image (if available)


And the highlight recovery shot (-1EV, could go further next time, maybe -1.3EV. -1.6EV would probably be overkill according to the blinkies)
1M294079ETTLBig.jpg
Subscribe to see EXIF info for this image (if available)


And a 100% crop in the sequence of the combined shot with adjustments made afterwards, the overexposed shot and the underexposed shot:
1M294079Comp.jpg
Subscribe to see EXIF info for this image (if available)


1M294079ETTR.jpg
Subscribe to see EXIF info for this image (if available)

1M294079ETTL.jpg
Subscribe to see EXIF info for this image (if available)


And a shadow pull up from the ETTL shot:
1M294079ETTLPull.jpg
Subscribe to see EXIF info for this image (if available)


Time to edit: Less than a minute in terms of effort, most of it being things that a macro would probably knock off half the time.
Time spent waiting: Around a minute or so on my desktop and with a 32 bit version of Photoshop which means a lot of opening, editing, purging all and saving before quitting, opening... to avoid memory problems.

I'm not sure where I can leave the full resolution file to play with, it is only a 11mb file but the excessive degree of dynamic range compression could be tweaked in that to personal preferences. The example posted has more illumination on the foliage than what I could see with my own eyes when I took the shot.
 
Last edited:

Latest threads

Links on this page may be to our affiliates. Sales through affiliate links may benefit this site.
Mu-43 is a fan site and not associated with Olympus, Panasonic, or other manufacturers mentioned on this site.
Forum post reactions by Twemoji: https://github.com/twitter/twemoji
Forum GIFs powered by GIPHY: https://giphy.com/
Copyright © Amin Forums, LLC
Top Bottom