Emulating the effect of an ND filter on a Panasonic GH5M2/GH6

SrMi wrote:

Interceptor121 wrote:

SrMi wrote:

- ND filter cannot improve image quality. It can only deteriorate it.

- Frame averaging improves image quality.

- A noisy image has always lower image quality than a "noise-free" image, regardless of the noise reduction.

- The difference in image quality is sometimes small, sometimes significant.

ND filter doesn’t do anything to image quality and if it is crap it ruins it

Any filter has the potential to decrease IQ. Long exposure increases fixed pattern noise.

No fixed pattern noise is always there and stacking actually aggravates it this is the reason why you shoot long exposure noise reduction by subtracting the dark frame you eliminate any fixed pattern and hot and cold pixels

mean stacking is the least effective method of noise reduction and destroys details because in very dark area or bright area it mixes everything

This does not match with common experience, AFAIK. I do not know why your tests show that behavior.

This has been discussed at length on astrophotography forums and posts and this is the reason currently it is not used for anything. For averaging bias frames or dark frames you can use median

you need to start from a good image

using stacking in camera as a noise reduction method or a way to enhance dynamic range is not worth it as post processing obtains better results due to the limitation of the camera

I disagree. Landscape photographers using Phase One’s in-camera frame averaging are very happy wit that feature. I am very happy with results obtained by Live ND.

Phase one manages 1024 shots so you no longer need an ND1000 here with 64 you will still need an ND16 to obtain an ND1000

In real situation for example a waterfall in daylight your starting point of a single exposure is 1/500. You are far away from 1/60 in order to get there you need to push the camera to f/16 and ISO 64 which both have huge issues more serious than using a filter

using stacking in software does provide the benefit that you can manipulate the entire data set but at the end this takes away time and there is a limit to how many exposures you can manage

but

more generally mean and median stacking are methods to create a slow exposure not proxy for something else

will post the median stacking example so we can all see hot that looks

So LiveND does not replace the most used filter ND1000 and that is because most scenes are far brighter than 1/60

As a method to improve image quality is a bad idea and should not be used

It appears that some users are not familiar with stacking methods and noise reduction so I thought of providing some examples unfortunately this was in fact the case where I did NOT have an ND filter I had my PL10-25mm if I recall

So this is the single frame as processed by lightroom the result is pretty good and there is frankly nothing to complain

Single Frame

This is the same shot as processed with DxO it obviously looks sharper and cleaner but DxO also corrects vignetting so the edges are brighter

DxO

This is the mean stacking of 64 frames if you look in the bottom right you will see it is not better than DxO and in the distance is worse the edges of the mountain are not as sharp and there are two people on top you can still see as it averaged the pixels

The water obviously looks as if it was a long exposure and so do the clouds

Mean Stacking 64 frames

Finally this is the same stack using median you will notice that it is sharper than mean stacking and there is only one little person on the mountain as it was there for most time and the other person was obliterated by the median

Median Stacking

My conclusion:

1. If I wanted to have the long exposure effect I would take the median because it looks better and removes objects better

2. The original shot in lightroom is actually ok and there is not much to complain about for 90% of users

3. As I have DxO my best shot of the sequence is number 2

4. I actually also run an HDR bracket on the same scene and that is my favourite of all because it has a wider luminance range. Yes you can overdo HDR but if you don't it remains the best tool for images like this as the image has far stronger impact and tones

HDR bracket

Dakotan wrote:

You have a camera, right? All these posts are all graphs and numbers, and not a single photo demonstrating what you are talking about. You say it's better than LIVE ND in another post also full of graphs and then adjusted graphs after you sort out median vs mean.

All most of us care about is the end product. Show a comparison image, or ANY image and then we can all judge. Until then, all the graphs you can supply are practically worthless.

I have posted the examples so you can pixel peep

I am sorry you cannot grasp the theory behind the process perhaps you can do some reading

SrMi wrote:

Interceptor121 wrote:

SrMi wrote:

Interceptor121 wrote:

SrMi wrote:

- ND filter cannot improve image quality. It can only deteriorate it.

- Frame averaging improves image quality.

- A noisy image has always lower image quality than a "noise-free" image, regardless of the noise reduction.

- The difference in image quality is sometimes small, sometimes significant.

ND filter doesn’t do anything to image quality and if it is crap it ruins it

Any filter has the potential to decrease IQ. Long exposure increases fixed pattern noise.

No fixed pattern noise is always there and stacking actually aggravates it this is the reason why you shoot long exposure noise reduction by subtracting the dark frame you eliminate any fixed pattern and hot and cold pixels

mean stacking is the least effective method of noise reduction and destroys details because in very dark area or bright area it mixes everything

This does not match with common experience, AFAIK. I do not know why your tests show that behavior.

This has been discussed at length on astrophotography forums and posts and this is the reason currently it is not used for anything. For averaging bias frames or dark frames you can use median

you need to start from a good image

using stacking in camera as a noise reduction method or a way to enhance dynamic range is not worth it as post processing obtains better results due to the limitation of the camera

I disagree. Landscape photographers using Phase One’s in-camera frame averaging are very happy wit that feature. I am very happy with results obtained by Live ND.

Phase one manages 1024 shots so you no longer need an ND1000 here with 64 you will still need an ND16 to obtain an ND1000

In real situation for example a waterfall in daylight your starting point of a single exposure is 1/500. You are far away from 1/60 in order to get there you need to push the camera to f/16 and ISO 64 which both have huge issues more serious than using a filter

using stacking in software does provide the benefit that you can manipulate the entire data set but at the end this takes away time and there is a limit to how many exposures you can manage

but

more generally mean and median stacking are methods to create a slow exposure not proxy for something else

will post the median stacking example so we can all see hot that looks

So LiveND does not replace the most used filter ND1000 and that is because most scenes are far brighter than 1/60

As a method to improve image quality is a bad idea and should not be used

Plenty of reputable sources disagree with that statement. I suggest that interested parties do their own research and tests and either agree with you or me.

Which reputable sources? have you got them?
There are none around that have compared all those things as I have looking at one thing in isolation doesn’t really work

the examples are in the main thread you can see the outcomes it is fairly easy for anyone

jrsforums wrote:

Interceptor121 wrote:

jrsforums wrote:

Dakotan wrote:

You have a camera, right? All these posts are all graphs and numbers, and not a single photo demonstrating what you are talking about. You say it's better than LIVE ND in another post also full of graphs and then adjusted graphs after you sort out median vs mean.

All most of us care about is the end product. Show a comparison image, or ANY image and then we can all judge. Until then, all the graphs you can supply are practically worthless.

There are lots of example articles & videos on the web….Google it….no need for more here. For example: https://fstoppers.com/education/using-long-exposures-without-help-neutral-density-filters-345545

A benefit of doing in post vs. in camera is you can adjust the effects by varying number of images used. Plus, in many cases, can handhold if using for noise reduction (not so much for ND as longer exposures) as post auto alignment is excellent.

When you can't see the wood from the trees it is easy to go and come out with statement like that person

Anyway just to give an example on how going and stacking frames is not a good idea unless required when you want to do a long exposure this is a comparison of the stack of 64 frames against a single frame treated with DxO Deep prime.

The whole idea that a primitive system like mean stacking all of a sudden is better than new noise reduction software is ludicrous

Here you can see that mean stacking of course smudges detail everywhere not just where things move the mountain and the bridge have less details

Another claim is that mean stacking all of a sudden is comparable to HDR and that the shadows are amazing well they are not really as in the example here DxO matches and beats it

So while it may be useful to use stacking if you don't have a strong enough filter with you to achieve the long exposure you want it is not true that you should be systematically stack exposure as it makes the images look amazing, it does not and it softens them

I was curious about those claims so I was looking for something that 'would be more subtle but equally effective than HDR' as suggested here. So I though let's give it a go and also get a bit of a long exposure effect (not desired)

Bracketed shot

Stacked shot

Single shot

I would argue that the single shot looks better than the stacked shot in the end and the HDR shot is in another category but that was a 3 stops gap bracket

I can see your examples. However, I have the feeling That it is an apples to oranges comparison. You used DXO routines (probably noise & sharpening AI routines under the covers…no sure as I have not used) and have not applied AI routines, such as Topaz denoise & sharpen, to the tiff. If you have not used them, I believe you would be surprised with the results.

Edit: how about a view of the ND image just processed through LR, without the DxO routines?

All examples are in the main thread. Besides it is not apple to oranges as mean staking is a noise reduction method it does not achieve any IQ improvement the opposite

Interceptor121 wrote:

jrsforums wrote:

Interceptor121 wrote:

jrsforums wrote:

Dakotan wrote:

You have a camera, right? All these posts are all graphs and numbers, and not a single photo demonstrating what you are talking about. You say it's better than LIVE ND in another post also full of graphs and then adjusted graphs after you sort out median vs mean.

All most of us care about is the end product. Show a comparison image, or ANY image and then we can all judge. Until then, all the graphs you can supply are practically worthless.

There are lots of example articles & videos on the web….Google it….no need for more here. For example: https://fstoppers.com/education/using-long-exposures-without-help-neutral-density-filters-345545

A benefit of doing in post vs. in camera is you can adjust the effects by varying number of images used. Plus, in many cases, can handhold if using for noise reduction (not so much for ND as longer exposures) as post auto alignment is excellent.

When you can't see the wood from the trees it is easy to go and come out with statement like that person

Anyway just to give an example on how going and stacking frames is not a good idea unless required when you want to do a long exposure this is a comparison of the stack of 64 frames against a single frame treated with DxO Deep prime.

The whole idea that a primitive system like mean stacking all of a sudden is better than new noise reduction software is ludicrous

Here you can see that mean stacking of course smudges detail everywhere not just where things move the mountain and the bridge have less details

Another claim is that mean stacking all of a sudden is comparable to HDR and that the shadows are amazing well they are not really as in the example here DxO matches and beats it

So while it may be useful to use stacking if you don't have a strong enough filter with you to achieve the long exposure you want it is not true that you should be systematically stack exposure as it makes the images look amazing, it does not and it softens them

I was curious about those claims so I was looking for something that 'would be more subtle but equally effective than HDR' as suggested here. So I though let's give it a go and also get a bit of a long exposure effect (not desired)

Bracketed shot

Stacked shot

Single shot

I would argue that the single shot looks better than the stacked shot in the end and the HDR shot is in another category but that was a 3 stops gap bracket

I can see your examples. However, I have the feeling That it is an apples to oranges comparison. You used DXO routines (probably noise & sharpening AI routines under the covers…no sure as I have not used) and have not applied AI routines, such as Topaz denoise & sharpen, to the tiff. If you have not used them, I believe you would be surprised with the results.

Edit: how about a view of the ND image just processed through LR, without the DxO routines?

All examples are in the main thread. Besides it is not apple to oranges as mean staking is a noise reduction method it does not achieve any IQ improvement the opposite

The primary point I was making was that with the ND filter you processed the image in DxO which does NR and sharpening using sophisticated AI routines.  From what you said, the stacked image did get NR from stacking, but no (sophisticated) sharpening.

btw…you may have tried Topaz ‘ages ago’, but it has really improved over that last few years.  There are a lot of comparisons of DxO & DeNoise which put them on par.  However, I am suggesting that the stacked images do not need sharpening and Topaz Sharpen AI can do a great job.  Personally, I would do in PS, so I could easily mask out the areas I continued to want blurred.(Topaz sharpen includes masking, but I find it easier in PS…probably just more experience with PS).

jrsforums wrote:

Interceptor121 wrote:

jrsforums wrote:

Interceptor121 wrote:

jrsforums wrote:

Dakotan wrote:

You have a camera, right? All these posts are all graphs and numbers, and not a single photo demonstrating what you are talking about. You say it's better than LIVE ND in another post also full of graphs and then adjusted graphs after you sort out median vs mean.

All most of us care about is the end product. Show a comparison image, or ANY image and then we can all judge. Until then, all the graphs you can supply are practically worthless.

There are lots of example articles & videos on the web….Google it….no need for more here. For example: https://fstoppers.com/education/using-long-exposures-without-help-neutral-density-filters-345545

A benefit of doing in post vs. in camera is you can adjust the effects by varying number of images used. Plus, in many cases, can handhold if using for noise reduction (not so much for ND as longer exposures) as post auto alignment is excellent.

When you can't see the wood from the trees it is easy to go and come out with statement like that person

Anyway just to give an example on how going and stacking frames is not a good idea unless required when you want to do a long exposure this is a comparison of the stack of 64 frames against a single frame treated with DxO Deep prime.

The whole idea that a primitive system like mean stacking all of a sudden is better than new noise reduction software is ludicrous

Here you can see that mean stacking of course smudges detail everywhere not just where things move the mountain and the bridge have less details

Another claim is that mean stacking all of a sudden is comparable to HDR and that the shadows are amazing well they are not really as in the example here DxO matches and beats it

So while it may be useful to use stacking if you don't have a strong enough filter with you to achieve the long exposure you want it is not true that you should be systematically stack exposure as it makes the images look amazing, it does not and it softens them

I was curious about those claims so I was looking for something that 'would be more subtle but equally effective than HDR' as suggested here. So I though let's give it a go and also get a bit of a long exposure effect (not desired)

Bracketed shot

Stacked shot

Single shot

I would argue that the single shot looks better than the stacked shot in the end and the HDR shot is in another category but that was a 3 stops gap bracket

I can see your examples. However, I have the feeling That it is an apples to oranges comparison. You used DXO routines (probably noise & sharpening AI routines under the covers…no sure as I have not used) and have not applied AI routines, such as Topaz denoise & sharpen, to the tiff. If you have not used them, I believe you would be surprised with the results.

Edit: how about a view of the ND image just processed through LR, without the DxO routines?

All examples are in the main thread. Besides it is not apple to oranges as mean staking is a noise reduction method it does not achieve any IQ improvement the opposite

The primary point I was making was that with the ND filter you processed the image in DxO which does NR and sharpening using sophisticated AI routines. From what you said, the stacked image did get NR from stacking, but no (sophisticated) sharpening.

btw…you may have tried Topaz ‘ages ago’, but it has really improved over that last few years. There are a lot of comparisons of DxO & DeNoise which put them on par. However, I am suggesting that the stacked images do not need sharpening and Topaz Sharpen AI can do a great job. Personally, I would do in PS, so I could easily mask out the areas I continued to want blurred.(Topaz sharpen includes masking, but I find it easier in PS…probably just more experience with PS).

Mean stacking does averages sharpness suffers it is a by product of the process

To have an idea of the comparison see the shots I put in the examples they cover all situations in a scenario where I used this technique and I did not have an ND

However I also took bracketed shots and those are actually my preferred outcome of the situation. I had other cases where I had ND1000 but then of course I did not bother stacking the shots and another situation where I did not have an ND filter strong enough but only needed 16 shots to have the 2.5" I was after

The main limitation of stacking in camera or post is that you have a limit of 1/60 or whatever the readout speed is. There are many situations where the scenes has a starting point at 1/500 f/4

When you put your ND1000 you get your 2 seconds. However if you had to rely on the camera internal process you needed to shoot at f/11 which results in loss of lens performance. Six stops is not enough for most real life situation in broad daylight so at the end you need a filter in your pocket regardless. I carry 4 1 CPL ND8 ND64 ND1000 normally and I can stack if I need to remove objects

Interceptor121 wrote:

This is the mean stacking of 64 frames if you look in the bottom right you will see it is not better than DxO and in the distance is worse the edges of the mountain are not as sharp and there are two people on top you can still see as it averaged the pixels

The water obviously looks as if it was a long exposure and so do the clouds

Mean Stacking 64 frames

The softening you're seeing here is a result of user error, either at capture time or in post-processing, not from the stacking methodology. Mean stacking is an ensemble averaging methodology; if there is no motion in the frame there's no mechanism available for mean stacking to smear information between pixels. Here's a properly controlled example of mean stacking to show the difference in resolution between a single exposure and a 64 image stack.

[L] 64 image stack with LiveND mean averaging vs [R] single image

whumber wrote:

Interceptor121 wrote:

This is the mean stacking of 64 frames if you look in the bottom right you will see it is not better than DxO and in the distance is worse the edges of the mountain are not as sharp and there are two people on top you can still see as it averaged the pixels

The water obviously looks as if it was a long exposure and so do the clouds

Mean Stacking 64 frames

The softening you're seeing here is a result of user error, either at capture time or in post-processing, not from the stacking methodology. Mean stacking is an ensemble averaging methodology; if there is no motion in the frame there's no mechanism available for mean stacking to smear information between pixels. Here's a properly controlled example of mean stacking to show the difference in resolution between a single exposure and a 64 image stack.

[L] 64 image stack with LiveND mean averaging vs [R] single image

As landscape scenes are not evenly illuminated and light changes during few seconds contrary to your studio test an area that had light at one point and had less at another will be averaged as result the image will become softer and or course there is grass on a mountain and wind blows etc so some edges are fixed some move that’s normal

likewise in a very dark area where values are closer to black the noise is averaged and not eliminated

this is why nobody uses mean stacking for any serious noise reduction other methods that don’t mix all the data they find are more effective

there is a difference between doing tests on a strawberry and shooting landscapes outdoors

median stacking that is not perfect tends to discard values at the edges not as good as other methods but works better

the frames are the same with the same process so user error (lol) is not part of that procedure only but also part of median I used the same identical stack

The proof of this is very easy shoot 64 images in a stack don't let the camera blend them take them yourself build and align the stack in photoshop and then duplicate once you have it as smart objects process one with mean and one with median

If you subtracts the stacks it looks like pitch black but it is not perform a division and you will see the difference right on the edges where either there was some movement that median killed or where the light somehow changed and median killed the edges

To some extent median stacking is even better than taking a single long exposure however because read noise goes up with the frames and snr in stacking is only improved with the square root it actually makes no sense to stack a scene except to remove unwanted objects a long exposure is always better under the assumption that thermal effect are the same in the two (which is not always confirmed)

When you introduce ND filters though you also attenuate signal so stacking has an advantage however considering that noise reduction software does achieve the 2.3 stops that stacking 64 frames does there is no issue in addition 90% you need 10 stops not 6 as most scenes are around 1/500 shutter unless you want to go to f/16 and garble your images using 1/60 in camera is not an option. This example was favourable as it is after sunset or thereabout

Interceptor121 wrote:

whumber wrote:

Interceptor121 wrote:

This is the mean stacking of 64 frames if you look in the bottom right you will see it is not better than DxO and in the distance is worse the edges of the mountain are not as sharp and there are two people on top you can still see as it averaged the pixels

The water obviously looks as if it was a long exposure and so do the clouds

Mean Stacking 64 frames

The softening you're seeing here is a result of user error, either at capture time or in post-processing, not from the stacking methodology. Mean stacking is an ensemble averaging methodology; if there is no motion in the frame there's no mechanism available for mean stacking to smear information between pixels. Here's a properly controlled example of mean stacking to show the difference in resolution between a single exposure and a 64 image stack.

[L] 64 image stack with LiveND mean averaging vs [R] single image

As landscape scenes are not evenly illuminated and light changes during few seconds contrary to your studio test an area that had light at one point and had less at another will be averaged as result the image will become softer and or course there is grass on a mountain and wind blows etc so some edges are fixed some move that’s normal

likewise in a very dark area where values are closer to black the noise is averaged and not eliminated

this is why nobody uses mean stacking for any serious noise reduction other methods that don’t mix all the data they find are more effective

there is a difference between doing tests on a strawberry and shooting landscapes outdoors

median stacking that is not perfect tends to discard values at the edges not as good as other methods but works better

the frames are the same with the same process so user error (lol) is not part of that procedure only but also part of median I used the same identical stack

The proof of this is very easy shoot 64 images in a stack don't let the camera blend them take them yourself build and align the stack in photoshop and then duplicate once you have it as smart objects process one with mean and one with median

If you subtracts the stacks it looks like pitch black but it is not perform a division and you will see the difference right on the edges where either there was some movement that median killed or where the light somehow changed and median killed the edges

To some extent median stacking is even better than taking a single long exposure however because read noise goes up with the frames and snr in stacking is only improved with the square root it actually makes no sense to stack a scene except to remove unwanted objects a long exposure is always better under the assumption that thermal effect are the same in the two (which is not always confirmed)

When you introduce ND filters though you also attenuate signal so stacking has an advantage however considering that noise reduction software does achieve the 2.3 stops that stacking 64 frames does there is no issue in addition 90% you need 10 stops not 6 as most scenes are around 1/500 shutter unless you want to go to f/16 and garble your images using 1/60 in camera is not an option. This example was favourable as it is after sunset or thereabout

Try the mean stack again without alignment and see what you get.

whumber wrote:

Interceptor121 wrote:

whumber wrote:

Interceptor121 wrote:

This is the mean stacking of 64 frames if you look in the bottom right you will see it is not better than DxO and in the distance is worse the edges of the mountain are not as sharp and there are two people on top you can still see as it averaged the pixels

The water obviously looks as if it was a long exposure and so do the clouds

Mean Stacking 64 frames

The softening you're seeing here is a result of user error, either at capture time or in post-processing, not from the stacking methodology. Mean stacking is an ensemble averaging methodology; if there is no motion in the frame there's no mechanism available for mean stacking to smear information between pixels. Here's a properly controlled example of mean stacking to show the difference in resolution between a single exposure and a 64 image stack.

[L] 64 image stack with LiveND mean averaging vs [R] single image

As landscape scenes are not evenly illuminated and light changes during few seconds contrary to your studio test an area that had light at one point and had less at another will be averaged as result the image will become softer and or course there is grass on a mountain and wind blows etc so some edges are fixed some move that’s normal

likewise in a very dark area where values are closer to black the noise is averaged and not eliminated

this is why nobody uses mean stacking for any serious noise reduction other methods that don’t mix all the data they find are more effective

there is a difference between doing tests on a strawberry and shooting landscapes outdoors

median stacking that is not perfect tends to discard values at the edges not as good as other methods but works better

the frames are the same with the same process so user error (lol) is not part of that procedure only but also part of median I used the same identical stack

The proof of this is very easy shoot 64 images in a stack don't let the camera blend them take them yourself build and align the stack in photoshop and then duplicate once you have it as smart objects process one with mean and one with median

If you subtracts the stacks it looks like pitch black but it is not perform a division and you will see the difference right on the edges where either there was some movement that median killed or where the light somehow changed and median killed the edges

To some extent median stacking is even better than taking a single long exposure however because read noise goes up with the frames and snr in stacking is only improved with the square root it actually makes no sense to stack a scene except to remove unwanted objects a long exposure is always better under the assumption that thermal effect are the same in the two (which is not always confirmed)

When you introduce ND filters though you also attenuate signal so stacking has an advantage however considering that noise reduction software does achieve the 2.3 stops that stacking 64 frames does there is no issue in addition 90% you need 10 stops not 6 as most scenes are around 1/500 shutter unless you want to go to f/16 and garble your images using 1/60 in camera is not an option. This example was favourable as it is after sunset or thereabout

Try the mean stack again without alignment and see what you get.

Firstly As I have explained the stack is identical if two methods produce different results this just shows that the method is impacting the results

You are an engineer does that sound logical? It does to me

Interceptor121 wrote:

whumber wrote:

Interceptor121 wrote:

whumber wrote:

Interceptor121 wrote:

This is the mean stacking of 64 frames if you look in the bottom right you will see it is not better than DxO and in the distance is worse the edges of the mountain are not as sharp and there are two people on top you can still see as it averaged the pixels

The water obviously looks as if it was a long exposure and so do the clouds

Mean Stacking 64 frames

The softening you're seeing here is a result of user error, either at capture time or in post-processing, not from the stacking methodology. Mean stacking is an ensemble averaging methodology; if there is no motion in the frame there's no mechanism available for mean stacking to smear information between pixels. Here's a properly controlled example of mean stacking to show the difference in resolution between a single exposure and a 64 image stack.

[L] 64 image stack with LiveND mean averaging vs [R] single image

As landscape scenes are not evenly illuminated and light changes during few seconds contrary to your studio test an area that had light at one point and had less at another will be averaged as result the image will become softer and or course there is grass on a mountain and wind blows etc so some edges are fixed some move that’s normal

likewise in a very dark area where values are closer to black the noise is averaged and not eliminated

this is why nobody uses mean stacking for any serious noise reduction other methods that don’t mix all the data they find are more effective

there is a difference between doing tests on a strawberry and shooting landscapes outdoors

median stacking that is not perfect tends to discard values at the edges not as good as other methods but works better

the frames are the same with the same process so user error (lol) is not part of that procedure only but also part of median I used the same identical stack

The proof of this is very easy shoot 64 images in a stack don't let the camera blend them take them yourself build and align the stack in photoshop and then duplicate once you have it as smart objects process one with mean and one with median

If you subtracts the stacks it looks like pitch black but it is not perform a division and you will see the difference right on the edges where either there was some movement that median killed or where the light somehow changed and median killed the edges

To some extent median stacking is even better than taking a single long exposure however because read noise goes up with the frames and snr in stacking is only improved with the square root it actually makes no sense to stack a scene except to remove unwanted objects a long exposure is always better under the assumption that thermal effect are the same in the two (which is not always confirmed)

When you introduce ND filters though you also attenuate signal so stacking has an advantage however considering that noise reduction software does achieve the 2.3 stops that stacking 64 frames does there is no issue in addition 90% you need 10 stops not 6 as most scenes are around 1/500 shutter unless you want to go to f/16 and garble your images using 1/60 in camera is not an option. This example was favourable as it is after sunset or thereabout

Try the mean stack again without alignment and see what you get.

Firstly As I have explained the stack is identical if two methods produce different results this just shows that the method is impacting the results

You are an engineer does that sound logical? It does to me

Humor me

Interceptor121 wrote:

I have a lot of experience in stacking as I do that for astrophotography. Mean and Median are primitive methods that are not used because they are not that good. Sigma clipping is what is used in various sauces but that is not available through photoshop

I missed this earlier. You may be interested to know that sigma clipping is a mean averaging method where we throw away all pixels that lie beyond one standard deviation of the median and then take the mean average of the remaining pixels.

whumber wrote:

Interceptor121 wrote:

whumber wrote:

Interceptor121 wrote:

whumber wrote:

Interceptor121 wrote:

This is the mean stacking of 64 frames if you look in the bottom right you will see it is not better than DxO and in the distance is worse the edges of the mountain are not as sharp and there are two people on top you can still see as it averaged the pixels

The water obviously looks as if it was a long exposure and so do the clouds

Mean Stacking 64 frames

The softening you're seeing here is a result of user error, either at capture time or in post-processing, not from the stacking methodology. Mean stacking is an ensemble averaging methodology; if there is no motion in the frame there's no mechanism available for mean stacking to smear information between pixels. Here's a properly controlled example of mean stacking to show the difference in resolution between a single exposure and a 64 image stack.

[L] 64 image stack with LiveND mean averaging vs [R] single image

As landscape scenes are not evenly illuminated and light changes during few seconds contrary to your studio test an area that had light at one point and had less at another will be averaged as result the image will become softer and or course there is grass on a mountain and wind blows etc so some edges are fixed some move that’s normal

likewise in a very dark area where values are closer to black the noise is averaged and not eliminated

this is why nobody uses mean stacking for any serious noise reduction other methods that don’t mix all the data they find are more effective

there is a difference between doing tests on a strawberry and shooting landscapes outdoors

median stacking that is not perfect tends to discard values at the edges not as good as other methods but works better

the frames are the same with the same process so user error (lol) is not part of that procedure only but also part of median I used the same identical stack

The proof of this is very easy shoot 64 images in a stack don't let the camera blend them take them yourself build and align the stack in photoshop and then duplicate once you have it as smart objects process one with mean and one with median

If you subtracts the stacks it looks like pitch black but it is not perform a division and you will see the difference right on the edges where either there was some movement that median killed or where the light somehow changed and median killed the edges

To some extent median stacking is even better than taking a single long exposure however because read noise goes up with the frames and snr in stacking is only improved with the square root it actually makes no sense to stack a scene except to remove unwanted objects a long exposure is always better under the assumption that thermal effect are the same in the two (which is not always confirmed)

When you introduce ND filters though you also attenuate signal so stacking has an advantage however considering that noise reduction software does achieve the 2.3 stops that stacking 64 frames does there is no issue in addition 90% you need 10 stops not 6 as most scenes are around 1/500 shutter unless you want to go to f/16 and garble your images using 1/60 in camera is not an option. This example was favourable as it is after sunset or thereabout

Try the mean stack again without alignment and see what you get.

Firstly As I have explained the stack is identical if two methods produce different results this just shows that the method is impacting the results

You are an engineer does that sound logical? It does to me

Humor me

Knock yourself off. What really strikes me of your flawed thinking is that you think taking averages would preserve detail a noise reduction software would not. Except taking averages IS noise reduction it is just a poor way of doing it.

Here the three examples first just the single frame top of the stack

Second the stack not aligned with mean (I have to say it improved the far objects but deteriorated the near objects)

And finally the median of the same not aligned stack if you look at certain details they are less muffled visible on the nearby rocks for example

However NONE of them does a better job of DxO on a single frame and in terms of sharpness none increases on a single untreated frame as you would expect

So in short although the measures that Bill has done and the theory indicate an improvement in noise reduction in effect this is just apparent in the really noisy areas and in normal circumstances doing this trick to improve matters does not achieve anything you cannot already do with a noise reduction software

If you wanted to have a long exposure effect though it works and for convenience use median stacking and do it in post so if a bird or someone walks in the frame like cow number 2 above is deleted.

The whole reason why I went out and shot almost a thousand frames to compare the various methods was because I was lead to believe that there was a visible improvement and even an alternative to HDR shots but there is not.

A chart does not actually measure in full how an image looks it is just one of many data points and neither does shooting a not moving on object in studio when this is a tool for landscape photography of which I do quite a bit recently

whumber wrote:

Interceptor121 wrote:

I have a lot of experience in stacking as I do that for astrophotography. Mean and Median are primitive methods that are not used because they are not that good. Sigma clipping is what is used in various sauces but that is not available through photoshop

I missed this earlier. You may be interested to know that sigma clipping is a mean averaging method where we throw away all pixels that lie beyond one standard deviation of the median and then take the mean average of the remaining pixels.

Are you joking now? Of course if you exclude pixels and do the mean of the remaining population you end up with a very different result that if you average the whole population

Please stop using google and insulting my intelligence I take a few thousands frame of night photography each year and I stack them

So there is no we there is just I do this and you read about it, I have yet to see an example where you actually did some work with stacked exposures?

Interceptor121 wrote:

whumber wrote:

Interceptor121 wrote:

I have a lot of experience in stacking as I do that for astrophotography. Mean and Median are primitive methods that are not used because they are not that good. Sigma clipping is what is used in various sauces but that is not available through photoshop

I missed this earlier. You may be interested to know that sigma clipping is a mean averaging method where we throw away all pixels that lie beyond one standard deviation of the median and then take the mean average of the remaining pixels.

Are you joking now? Of course if you exclude pixels and do the mean of the remaining population you end up with a very different result that if you average the whole population

You only end up with a very different result if you have large outliers. You're claiming that mean averaging fundamentally reduces detail, this is simply not true. Are you now claiming that mean averaging only reduces detail if we don't throw out outliers beyond the standard deviation?

Please stop using google and insulting my intelligence I take a few thousands frame of night photography each year and I stack them

I believe that you take thousands of frames of night photography every year. I don't believe that you know what you are doing. You seem to believe that a single dark frame eliminates read noise and thermal noise, you didn't even know enough to realize that auto-aligning a static scene would screw up the stacking, and I'm pretty sure you don't understand why the median stacking handles it better.

whumber wrote:

Interceptor121 wrote:

whumber wrote:

Interceptor121 wrote:

I have a lot of experience in stacking as I do that for astrophotography. Mean and Median are primitive methods that are not used because they are not that good. Sigma clipping is what is used in various sauces but that is not available through photoshop

I missed this earlier. You may be interested to know that sigma clipping is a mean averaging method where we throw away all pixels that lie beyond one standard deviation of the median and then take the mean average of the remaining pixels.

Are you joking now? Of course if you exclude pixels and do the mean of the remaining population you end up with a very different result that if you average the whole population

You only end up with a very different result if you have large outliers. You're claiming that mean averaging fundamentally reduces detail, this is simply not true. Are you now claiming that mean averaging only reduces detail if we don't throw out outliers beyond the standard deviation?

Of course it does but that happens to long exposures too is part of the process. It would only not do if everything was perfectly still and subject to constant illumination like your strawberry example because the average of n identical samples is equal to that sample

when you look at situations of high variability and high noise you can have very different vaiues over 64 samples on a scale of 4000

but that is basic

Please stop using google and insulting my intelligence I take a few thousands frame of night photography each year and I stack them

I believe that you take thousands of frames of night photography every year

except i have good shots and you don’t

. I don't believe that you know what you are doing. You seem to believe that a single dark frame eliminates read noise and thermal noise,

i don’t and not even a set of dark frame eliminates noise actually but that’ doesn’t matter to you as you dont take those shots anyway am i right?

you didn't even know enough to realize that auto-aligning a static scene would screw up the stacking, and I'm pretty sure you don't understand why the median stacking handles it

Auto aligning a scene doesn’t screw up the stacking and doesn’t change the conclusions

now when are you showing any pictures you take? Have you ever taken s stacked shot outside your kitchen?

Funny how the internet gives power to argue to someone with zero practical experience