Canon 100D on 72 ED-R, iso 400 and 800 comparison

[StargeezerTim]

Well I obviously have nothing better to do so I have done a comparison between outcomes for standardised processing on a 400 iso and 800 iso image. Up till now, I have tended to use iso 1600 but recently thought i'd experiment with iso 400 and it was OK, hence the comparison to help me decided iso on a more rational basis.

Each image is 9 X 600 secs, UHC filter, 400 iso captured first, then the 800 iso in the same run, e.g. 2 X 90 min runs, same night, same guiding etc. Started imaging in astro dark and moon didn't effect either set of subs. They were stacked in DSS, Median Kappa Sigma mode, no tweaking of autosave files, just saved and converted to 16 bit images in PS. Here are the stacked images with no processing barring a 25 pt gap was put between the black point and the far left of the Histogram.

400 iso

800 iso

I then applied the same processing to both images. I did two medium strength curves and one mild strength curve on both images. After each curve application, I kept a 25 point gap between the black point and the far left of the histogram. The curves are saved as presets so there is no chance of applying different stretching to either image. Here are the images after stretching.

iso 400

iso 800

 

Finally, here is crop of the central part of both images so the differences are easier to see.

iso 400

iso 800

Conclusions. I was surprised that the 800 iso image, even without stretching, seemed to exaggerate the star size and brightness. This is most clearly seen on the processed crops, where the stars on the 800 image are actually blown, it seems to me. So for images of 10 min sub length, processing on the image needs to be very gentle and steps taken to protect the stars very early on in the process (e.g. saving a star mask). The iso 800 image did not capture any 'extra' data, as far as I can see, though the fainter surrounding nebulosity was easier to see at these levels of processing than on the iso 400 image. I think that iso 400 is probably the better setting for 10 min subs, fainter surrounding nebulosity can be brought out whilst masking the rest but I stand  a better chance of getting nice star sizes and colour with iso 400 than 800, I think!

The iso 800 image is also much noisier though I have plenty of practice in dealing with that! 

I know that you cant generalise from one example, even though I tried to control the process, but as a real world example, I have found it useful, at least. Whether the same differences occur when it the temperature is much colder, who knows? Also, the picture may be different with shorter subs as well.

If anyone has any thoughts or observations about the results or process, I look forward to your comments.... 

 

 

 

[D4N]

The 400 certainly looks better at that level of processing, I imagine your stars would have been huge if you had used 1600!

[wxsatuser]

At ISO 800 the well depth would be approx half that at ISO 400.
If you use the same exposure time 800 is going towards blow out.

What you should do is half the 800 exposure and double the subs.

 

[sharkmelley]

The ISO 800 image looks brighter overall and that has the effect of both making the noise more apparent and of blowing out the stars.  It's not clear to me which step or steps in the processing sequence did this.  With more careful processing on the ISO 800 image you could make it look identical to the ISO 400 image except that a few of the stars would have cores that have saturated earlier.   There's no reason why stars in the ISO 800 image should appear larger.

Mark

[StargeezerTim]

21 minutes ago, sharkmelley said:

The ISO 800 image looks brighter overall and that has the effect of both making the noise more apparent and of blowing out the stars.  It's not clear to me which step or steps in the processing sequence did this.  With more careful processing on the ISO 800 image you could make it look identical to the ISO 400 image except that a few of the stars would have cores that have saturated earlier.   There's no reason why stars in the ISO 800 image should appear larger.

Mark

Just tried adjusting brightness and, yes, the star size seems larger the brighter the image. The 800 is much noisier though, even with brightness matched.

 

[sharkmelley]

3 minutes ago, StargeezerTim said:

Just tried adjusting brightness and, yes, the star size seems larger the brighter the image. The 800 is much noisier though, even with brightness matched.

 

The ISO 800 image shouldn't be noisier unless the skyfog was brighter or the thermal noise was substantially greater.  To perform a fair comparison between your two images, a different approach may be required.  Note that the pixel values in the ISO 800 stacked image will be more or less double the pixel values in the ISO 400 stacked image.  So for the ISO 800 processing, try reducing the stacked image pixel values by 50% and then apply exactly the same processing operations to both images.  The end result should be that the images are virtually indistinguishable.

There is actually a good argument for using a higher ISO: higher ISOs generally reduce the fixed pattern noise and banding effects caused by the downstream electronics that are often seen in the image background with Canon sensors - see http://www.clarkvision.com/articles/iso/ for example.

Mark

[StargeezerTim]

12 minutes ago, sharkmelley said:

 Note that the pixel values in the ISO 800 stacked image will be more or less double the pixel values in the ISO 400 stacked image.  So for the ISO 800 processing, try reducing the stacked image pixel values by 50% and then apply exactly the same processing operations to both images.

What do you mean by 'pixel values' and how can this be changed in photoshop?

[sharkmelley]

14 minutes ago, StargeezerTim said:

What do you mean by 'pixel values' and how can this be changed in photoshop?

By pixel values I mean the R,G and B values in each point of your image.  I think to scale them to 50% in PS you need to apply a "Levels" operation and in the Levels dialog box set the output levels range to be 0-128 instead of 0-255.   There's probably a much better way - I'm not really that familiar with PS.

Mark

[StargeezerTim]

1 hour ago, sharkmelley said:

By pixel values I mean the R,G and B values in each point of your image.  I think to scale them to 50% in PS you need to apply a "Levels" operation and in the Levels dialog box set the output levels range to be 0-128 instead of 0-255.   There's probably a much better way - I'm not really that familiar with PS.

Mark

This seems to work Mark. I'm flabbergasted! If anything, the higher iso now has less noise than the 400 iso. See image, the right side is the iso 800 and the left the iso 400. Conditions were actually better for the iso 800 capture.

Is there a formula or something when different iso's are used. Do I reduce output levels by a further half for 1600 iso? Are there any drawbacks to doing this? If I can get the same image with same noise at iso 400 or 800, which one should i choose, or is 1600 as good or better? Sorry for asking so many Q's but I'm actually a bit confused 

[sharkmelley]

33 minutes ago, StargeezerTim said:

This seems to work Mark. I'm flabbergasted! If anything, the higher iso now has less noise than the 400 iso. See image, the right side is the iso 800 and the left the iso 400. Conditions were actually better for the iso 800 capture.

Is there a formula or something when different iso's are used. Do I reduce output levels by a further half for 1600 iso? Are there any drawbacks to doing this? If I can get the same image with same noise at iso 400 or 800, which one should i choose, or is 1600 as good or better? Sorry for asking so many Q's but I'm actually a bit confused 

 

I'm glad it worked.   The Sensorgen results for your camera ( http://www.sensorgen.info/CanonEOS-100D.html) indicate that read noise drops noticeably from ISO 400 to ISO 800 and then again to ISO 1600.  This might be the effect you are seeing or, as you say, it could be the improved sky conditions.  It is not easy to quantify the sources of noise in your image without looking at raw lights, darks and bias files. 

Yes you would reduce output levels by half again for each doubling of the ISO.  However, in general it is not a good idea to reduce levels as a first processing step - I suggested it only as a quick way to obtain a fair comparison.  It is better to tweak your processing slightly to compensate for the histogram moved to the right.  The only problem with using a higher ISO is that the brighter stars will saturate earlier because of reduced dynamic range.  But you might feel that is a price worth paying if it results in reduced background noise levels.  It's very much an individual choice - there are no rights or wrongs here.

Mark

[Adam J]

I did this same experiment a few months back. This  I always tell people not to use ISO 1600 as its significantly worse then either 400 or 800 (at least for canons in the last 8 years), the selection of 400 or 800 will be camera model dependent. People tend to get stove piped into looking at the read noise without realising that the amplifier induced noise increases in a non linear manor with gain. 

On 15/09/2016 at 20:09, sharkmelley said:

The ISO 800 image looks brighter overall and that has the effect of both making the noise more apparent and of blowing out the stars.  It's not clear to me which step or steps in the processing sequence did this.  With more careful processing on the ISO 800 image you could make it look identical to the ISO 400 image except that a few of the stars would have cores that have saturated earlier.   There's no reason why stars in the ISO 800 image should appear larger.

Mark

It looks brighter because it started out brighter due to the higher gain an identical stretch being used as with the ISO 400 image will result in the two images being at different average brightnesses. If you want to do a real comparison they need to be stretched so that the peak of the histogram is in the same place for both images. Visually you will see less noise in the ISO 400 image as although the stretch was identical its has still not ended up at equal brightness to the ISO 800 image. 

On 15/09/2016 at 20:48, sharkmelley said:

The ISO 800 image shouldn't be noisier unless the skyfog was brighter or the thermal noise was substantially greater.  To perform a fair comparison between your two images, a different approach may be required.  Note that the pixel values in the ISO 800 stacked image will be more or less double the pixel values in the ISO 400 stacked image.  So for the ISO 800 processing, try reducing the stacked image pixel values by 50% and then apply exactly the same processing operations to both images.  The end result should be that the images are virtually indistinguishable.

There is actually a good argument for using a higher ISO: higher ISOs generally reduce the fixed pattern noise and banding effects caused by the downstream electronics that are often seen in the image background with Canon sensors - see http://www.clarkvision.com/articles/iso/ for example.

Mark

You are correct about the brightness, however it is not true to say that it should not be nosier. Additional noise is added by the thermal noise within amplifier itself during amplification. Internal amp noise will increase with amplification level in a non linear fashion, as the ISO is increased overall noise initially improves as the amplification of the signal 'overcomes' the read noise, however after a certain point this reverses and total noise increases as the contribution of the thermal noise within the amplifier becomes the limiting factor instead of the read noise. Exposure length plays a important part in where this balance lays. As such for very short exposures a higher ISO is required to overcome read noise as the integrated thermal noise component is less significant. But for 600s exposures you are going to be limited by the various sources of thermal noise (there is more than just one source).

To have a true comparison of noise within the image you need to match the average brightness between the images (something that is not achieved by performing an identical stretch).  However to evaluate noise digitally all you need to do is examine the standard deviation of the background pixels as this is independent of the average pixel value. 

In my own camera I find that:

1) Signal to noise about twice as good for 2400s @ ISO400 vs 1200s @ ISO800 (hence there is not much in it).

2) Signal to noise with 1200s @ ISO 800 is more than twice as good as 600s @ ISO 1600 (hence moving from ISO 800 to ISO 1600 gives a negative return for longer exposures). 

But that is for single exposures....by its nature you will get twice the number of exposures for half the imaging time over a fixed length imaging session and if you dither between exposures stacking can make up for the difference in noise in the shorted higher ISO exposures.... 

If you keep the exposure length the same and change the ISO chances are that for longer individual exposures you are failing to fully exploit the reduction in the thermal noise inherent within the amplifier and so the difference between the two ISO settings will be less pronounced. 

In the end there are so many factors to consider when making this particular comparison that it causes endless debate because you normally find people are never comparing apples with apples as factors external to the camera such as sky glow and temperature are constantly shifting the balance as well as the balance of the various noise sources within the camera.

Having said this, I have personally concluded that for most cameras in most conditions (with some notable exceptions) you will do well with either ISO 400 or ISO 800, where as ISO 1600 is a little more niche and mainly beneficial when exposure length is limited in a non guided set-up.      

 

[dph1nm]

1 hour ago, Adam J said:

1) Signal to noise about twice as good for 2400s @ ISO400 vs 1200s @ ISO800 (hence there is not much in it).

2) Signal to noise with 1200s @ ISO 800 is more than twice as good as 600s @ ISO 1600 (hence moving from ISO 800 to ISO 1600 gives a negative return for longer exposures). 

Hmm - s/n should increase by sqrt(2) not 2 when you double the exposure (unless, of course, you are read-noise limited) ...

NIgelM

[sharkmelley]

3 hours ago, Adam J said:

Internal amp noise will increase with amplification level in a non linear fashion, as the ISO is increased overall noise initially improves as the amplification of the signal 'overcomes' the read noise, however after a certain point this reverses and total noise increases as the contribution of the thermal noise within the amplifier becomes the limiting factor instead of the read noise.

 

I've not heard of such an effect.  If such an effect does exist, surely it would be included in the read noise estimates ( e.g at Sensorgen) and one would see read noise estimates increasing with ISO.

 

1 hour ago, dph1nm said:

Hmm - s/n should increase by sqrt(2) not 2 when you double the exposure (unless, of course, you are read-noise limited) ...

Agreed.

Mark

[Adam J]

I cool my DSLR to around -5c to -10c dependent on ambient temperatures so I suppose its possible that I am still read noise limited even in very long exposures.

[sharkmelley]

4 hours ago, Adam J said:

I cool my DSLR to around -5c to -10c dependent on ambient temperatures so I suppose its possible that I am still read noise limited even in very long exposures.

Unlikely, unless you are using a slow F-ratio under pristine skies (or doing narrowband imaging).  I'm guessing you already know how to work out if you are read noise limited but if not, it's very easy to do a quick calculation.  Open one of your light frames and let us know:

• the model of camera
• the ISO used
• the background sky level in ADU (i.e the RGB values )
• any narrowband filter being used

We can then make a reasonably good estimate of  the noise due to sky fog and compare it with the read noise - all using easily available data. 

That won't give us the thermal noise but there is also a quick way that can be measured:

• take two successive dark frames towards the end of the imaging run
• subtract one from the other (with an added offset to prevent any negative or clipped numbers)
• calculate the standard deviation

Then knowing the camera model and ISO we can again make a reasonably good estimate.

In my opinion, read noise, thermal noise and sky fog noise are 3 crucial pieces of data that every serious imager should want to know in order to work out optimum exposure length (and optimum ISO for DSLR imaging).

Mark

[Adam J]

4 hours ago, sharkmelley said:

Unlikely, unless you are using a slow F-ratio under pristine skies (or doing narrowband imaging).  I'm guessing you already know how to work out if you are read noise limited but if not, it's very easy to do a quick calculation.  Open one of your light frames and let us know:

• the model of camera
• the ISO used
• the background sky level in ADU (i.e the RGB values )
• any narrowband filter being used

We can then make a reasonably good estimate of  the noise due to sky fog and compare it with the read noise - all using easily available data. 

That won't give us the thermal noise but there is also a quick way that can be measured:

• take two successive dark frames towards the end of the imaging run
• subtract one from the other (with an added offset to prevent any negative or clipped numbers)
• calculate the standard deviation

Then knowing the camera model and ISO we can again make a reasonably good estimate.

In my opinion, read noise, thermal noise and sky fog noise are 3 crucial pieces of data that every serious imager should want to know in order to work out optimum exposure length (and optimum ISO for DSLR imaging).

Mark

I am doing narrow band imaging with 2" Baader 7nm H-a and Baader 8.5nm OIII filters. Also I use a Canon 1000D so its read noise is quite prominent with a vertical banding towards the right hand side of the image. I have a second 1000D sensor on the table at home right now with the cover glass removed ready to attempt a mono mod. I have only been using narrow band for a couple of months since I got my cooling up and running. 

I can normally still visually discern the pattern noise mixed in with the thermal noise in a 600s dark frame when the camera is cooled. So the thermal noise has definitely not swamped the read noise completely, and since they started to turn the street lights out at 11:30 around me sky glow is definitely not limiting me with the narrow band filters, it is starting to limit me at about 10 - 20 mins with the CLS filter. After work ill pull out some dark frames to show you and let you play with....not that I normally use dark frames much, just bias, my cooling is not regulated and dark frames often make things worse for me as the temperature tends to drop over the imaging session.  

[dph1nm]

3 hours ago, Adam J said:

Also I use a Canon 1000D so its read noise is quite prominent with a vertical banding towards the right hand side of the image.

Mine has this at any ISO below 1600, but 1600 is clean. I believe (i.e. I read a long time ago in another forum) that Canon use different circuitry for ISO1600 on this camera, hence the change in read noise behaviour.

NIgelM

p.s. depends on pixel size and aperture, but from a dark site, and especially with narrow band filters, it can take a surprisingly long time to overcome read noise.

[Adam J]

2 hours ago, dph1nm said:

Mine has this at any ISO below 1600, but 1600 is clean. I believe (i.e. I read a long time ago in another forum) that Canon use different circuitry for ISO1600 on this camera, hence the change in read noise behaviour.

NIgelM

p.s. depends on pixel size and aperture, but from a dark site, and especially with narrow band filters, it can take a surprisingly long time to overcome read noise.

Yes there is less pattern noise at 1600, but if you are cooling to -10 you can still see just about see it. That says to me that its being swamped by the noise introduced when moving to 1600. Makes some sense as although the noise at 1600 appears to lack the pattern component the standard deviation is very much worse than at 800. Personally I would take the fixed pattern noise over the random noise any day as its easy to remove it with bias frames.

I know Canon DSLR have a duel amp setup and that the ISO is normally controlled by the first amp in the series. I would not mind betting that the first amp is maxed out at ISO 800 and they get to ISO 1600 by increasing the amplification on the second amp and I would not mind betting that the second amp is nosier than the first.