Noise Reduction in Stacked Images (a bit technical)

Started 1 month ago | Discussions
bclaff Veteran Member • Posts: 6,002
Noise Reduction in Stacked Images (a bit technical)
7

My post on ILCE-7RM3 Pixel Shift Multi Shooting (PSMS) Photographic Dynamic Range (PDR) at PhotonsToPhotos  shows this chart:

and states you don't get the noise reduction you might expect because "read noise is incurred on all frames"

But I misspoke, it's not the read noise but the Fixed Pattern Noise (FPN) that sets a limit on the noise reduction from stacking.

So I went back to my FPN test files and produced this chart:

Black noise is the lower curve using the left y-axis and illuminated noise is the upper curve using the right y-axis.

The chart shows noise rather than stops. After 4 images black and illuminated read noise had dropped about 1.0 and 0.7 stops respectively. (Note that PDR shows about 1.15 stops)

If you could stack an infinite number of images the black and illuminated improvements would be about 4.1 and 1.3 stops respectively because no matter how many images you stack you can't eliminate the FPN.

In fact, at some point, Image Quality (IQ) might worsen rather than improve because stacking can make FPN more apparent.
You can see that effect in the PhotonsToPhotos Sensor Heatmap visualization:

False colors are not important they simply help you see patterns if any.

(I think the stacked illuminated image looks worse than the single.)
So, stacking to improve Image Quality (IQ) is worthwhile but dynamic range improvements are limited by Fixed Pattern Noise (FPN) and stacking too many images may visibly expose FPN and lower IQ.

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Astrophotographer 10 Veteran Member • Posts: 9,979
Re: Noise Reduction in Stacked Images (a bit technical)

bclaff wrote:

My post on ILCE-7RM3 Pixel Shift Multi Shooting (PSMS) Photographic Dynamic Range (PDR) at PhotonsToPhotos shows this chart:

and states you don't get the noise reduction you might expect because "read noise is incurred on all frames"

But I misspoke, it's not the read noise but the Fixed Pattern Noise (FPN) that sets a limit on the noise reduction from stacking.

So I went back to my FPN test files and produced this chart:

Black noise is the lower curve using the left y-axis and illuminated noise is the upper curve using the right y-axis.

The chart shows noise rather than stops. After 4 images black and illuminated read noise had dropped about 1.0 and 0.7 stops respectively. (Note that PDR shows about 1.15 stops)

If you could stack an infinite number of images the black and illuminated improvements would be about 4.1 and 1.3 stops respectively because no matter how many images you stack you can't eliminate the FPN.

In fact, at some point, Image Quality (IQ) might worsen rather than improve because stacking can make FPN more apparent.
You can see that effect in the PhotonsToPhotos Sensor Heatmap visualization:

False colors are not important they simply help you see patterns if any.

(I think the stacked illuminated image looks worse than the single.)
So, stacking to improve Image Quality (IQ) is worthwhile but dynamic range improvements are limited by Fixed Pattern Noise (FPN) and stacking too many images may visibly expose FPN and lower IQ.

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Good to know Bill.

Do you have an estimate of when its too many being stacked? I stack images in nightscapes at high ISO.

Greg.

 Astrophotographer 10's gear list:Astrophotographer 10's gear list
Sony Alpha a7R II Sony Alpha a7R III Sony FE 55mm F1.8 Zeiss Batis 85mm F1.8 Zeiss Loxia 21mm F2.8 +1 more
OP bclaff Veteran Member • Posts: 6,002
Re: Noise Reduction in Stacked Images (a bit technical)

Astrophotographer 10 wrote:

bclaff wrote:

...

Do you have an estimate of when its too many being stacked? I stack images in nightscapes at high ISO.

There is no hard and fast rule; only that you can have too much of a good thing

I'm sure you would notice if you have gone too far with your stacking.

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

OP bclaff Veteran Member • Posts: 6,002
Re: Noise Reduction in Stacked Images (a bit technical)
1

Some people might find the results easier to interpret as stops:

It looks like if you're trying to improve shadows that 8-12 images would be enough and things really level off after that.

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

AudiiDudii Regular Member • Posts: 284
Re: Noise Reduction in Stacked Images (a bit technical)

bclaff wrote:

Some people might find the results easier to interpret as stops:

It looks like if you're trying to improve shadows that 8-12 images would be enough and things really level off after that.

I possess no experise in these matters, but when doing my urban night photography, I take seven exposures with my A7R then median-blend them in CS6 during post-processing.

I find seven exposures to be a good compromise between the time spent taking the photos (my typical exposure is between 15-30 seconds, so seven exposures takes me close to three minutes total) and reducing noise / recovering shadow details. Plus, by using an odd number of files, there's no need for Photoshop to average any pixels to determine their median value, as I presume would be the case if I fed it an even number of files.

JG

OP bclaff Veteran Member • Posts: 6,002
Re: Noise Reduction in Stacked Images (a bit technical)

AudiiDudii wrote:

bclaff wrote:

Some people might find the results easier to interpret as stops:

It looks like if you're trying to improve shadows that 8-12 images would be enough and things really level off after that.

I possess no experise in these matters, but when doing my urban night photography, I take seven exposures with my A7R then median-blend them in CS6 during post-processing.

I find seven exposures to be a good compromise between the time spent taking the photos (my typical exposure is between 15-30 seconds, so seven exposures takes me close to three minutes total) and reducing noise / recovering shadow details. Plus, by using an odd number of files, there's no need for Photoshop to average any pixels to determine their median value, as I presume would be the case if I fed it an even number of files.

Seven is a nice number Of course median blending is not quite the same as averaging.
I checked my figures and the A7R behaves very much like the A7RM3 in regards to stacking in the shadows.
Both get about 1.4 stops improvement with 7 images.

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

AudiiDudii Regular Member • Posts: 284
Re: Noise Reduction in Stacked Images (a bit technical)

bclaff wrote:

I checked my figures and the A7R behaves very much like the A7RM3 in regards to stacking in the shadows.
Both get about 1.4 stops improvement with 7 images.

An improvement of 1.4 stops matches my observed results fairly closely.  Thanks for going back to your A7R data to confirm this for me. 

JG

OP bclaff Veteran Member • Posts: 6,002
Re: Noise Reduction in Stacked Images (a bit technical)

AudiiDudii wrote:

bclaff wrote:

I checked my figures and the A7R behaves very much like the A7RM3 in regards to stacking in the shadows.
Both get about 1.4 stops improvement with 7 images.

An improvement of 1.4 stops matches my observed results fairly closely. Thanks for going back to your A7R data to confirm this for me.

Thanks for the anecdotal confirmation that the science actually works

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Astrophotographer 10 Veteran Member • Posts: 9,979
Re: Noise Reduction in Stacked Images (a bit technical)

bclaff wrote:

AudiiDudii wrote:

bclaff wrote:

I checked my figures and the A7R behaves very much like the A7RM3 in regards to stacking in the shadows.
Both get about 1.4 stops improvement with 7 images.

An improvement of 1.4 stops matches my observed results fairly closely. Thanks for going back to your A7R data to confirm this for me.

Thanks for the anecdotal confirmation that the science actually works

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

So at 7 images you are at 1.4 stops which sounds from the original post to be max improvement (at least for illuminated noise not black noise).?

Greg.

 Astrophotographer 10's gear list:Astrophotographer 10's gear list
Sony Alpha a7R II Sony Alpha a7R III Sony FE 55mm F1.8 Zeiss Batis 85mm F1.8 Zeiss Loxia 21mm F2.8 +1 more
OP bclaff Veteran Member • Posts: 6,002
Re: Noise Reduction in Stacked Images (a bit technical)

Astrophotographer 10 wrote:

bclaff wrote:

AudiiDudii wrote:

bclaff wrote:

I checked my figures and the A7R behaves very much like the A7RM3 in regards to stacking in the shadows.
Both get about 1.4 stops improvement with 7 images.

An improvement of 1.4 stops matches my observed results fairly closely. Thanks for going back to your A7R data to confirm this for me.

Thanks for the anecdotal confirmation that the science actually works

So at 7 images you are at 1.4 stops which sounds from the original post to be max improvement (at least for illuminated noise not black noise).?

Sure. 1.4 stops of black improvement at 7 images is about the same as 1.3 stops maximum improvement for the illuminated.

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

sharkmelley
sharkmelley Contributing Member • Posts: 944
Calibration frames

The limitation imposed by the FPN is one reason that astrophotographers use calibration frames (flats, darks, bias) in their processing.  The calibration frames allow the effects of the FPN to be removed.

Mark

-- hide signature --

Takahashi Epsilon 180ED
H-alpha modified Sony A7S
http://www.markshelley.co.uk/Astronomy/

 sharkmelley's gear list:sharkmelley's gear list
Sony Alpha a7S +1 more
Keyboard shortcuts:
FForum MMy threads