Jump to content


Photo

Profiling Long-Exposure Performance of Canon DSLRs

  • Please log in to reply
13 replies to this topic

#1 Charlie Hein

Charlie Hein

    Voyager 1

  • *****
  • Administrators
  • Posts: 12359
  • Joined: 02 Nov 2003
  • Loc: 26.06.08N, +80.23.08W

Posted 02 July 2012 - 08:05 PM

Profiling the Long-Exposure Performance of a Canon DSLR

By Craig Stark

#2 piaras

piaras

    Viking 1

  • -----
  • Posts: 827
  • Joined: 26 Jan 2009
  • Loc: Niagara Region

Posted 02 July 2012 - 09:28 PM

Very interesting, now I will have to try longer exposures at lower ISOs.
Can not do anything about the internal changes to the file so that is a point to not worry about. Just like the chips that are semi linear with slight leakage and other issues, can not do anything about that either.

#3 chboss

chboss

    Viking 1

  • -----
  • Posts: 731
  • Joined: 24 Mar 2008
  • Loc: Zurich Switzerland

Posted 03 July 2012 - 12:06 AM

Very interesting article, that answers some observations I made when comparing darks from my DSLR...
Great work Craig!

regards
Chris

#4 rainycityastro

rainycityastro

    Viking 1

  • *****
  • Posts: 747
  • Joined: 29 Mar 2010
  • Loc: Seattle, WA

Posted 03 July 2012 - 03:43 PM

This is a really illuminating article! Makes me feel better about the money I recently spent on a QSI camera.
Craig, you do mention that you've tested other more modern DSLRs from canon (specifically the 5d2).
Do the 5d2 and other modern cameras also show similar behavior, i.e. longer exposures leading to lower dark current with a concomitant increase in dark noise? Does your advice on keeping the ISO low (around 400) apply to 5d2 as well?

#5 Craig

Craig

    Vendor (Stark Labs)

  • *****
  • Vendors
  • Posts: 768
  • Joined: 16 Sep 2003
  • Loc: Irvine, CA

Posted 04 July 2012 - 12:25 PM

The 5D II did show an increase in dark current with exposure duration, but I'm still pretty sure it's compensating - just not as much / overcompensating:
http://www.stark-lab...DSLRs_ATT09.pdf

In Table 1 there, multiply the system gain by 4 to get it back to 14-bit-actual-ADU-numbers (or look for something just under 0.25). The 50D would also then have 400 be the best starting point for ISO and the 5D II would be 800.

Craig

#6 themos

themos

    Sputnik

  • -----
  • Posts: 40
  • Joined: 01 Apr 2008

Posted 05 July 2012 - 04:49 AM

Dear Craig,

Back in October 2009, I followed the advice given in your earlier articles about measuring gain and read-noise and got some numbers for my 450D/XSi.

My camera's gain is 0.568 e-/ADU at ISO 400.
My camera's read error is 6.3 e- at ISO 400.

So I already "knew" that ISO 400 seemed to be the best setting. The thing is there is so much peer pressure to use the higher settings that I ended up going for ISO 800 for most of my shots since then. That's it, no more!

Looking at my notes from back then, I was puzzled by the dark frames too! I wrote

"The dark frames seem to contain values much less than the bias (1024), sometimes going down to 522 for the 20-min. It's not so much a dark current as a dark chaos."

#7 Jb32828

Jb32828

    Messenger

  • *****
  • Posts: 479
  • Joined: 01 Aug 2010
  • Loc: Orlando, FL USA

Posted 05 July 2012 - 07:43 AM

This is another excellent article that you have submitted. Thanks for the explanation. Intuitively, I have always had better actual data to work with the lower the ISO and longer the exposure, just never could figure out why.

Here is my question. Do you think that the "correction" of the image is done in firmware or software? Would simply programming some new software for the camera fix it? Have you tried the green lantern software?

#8 Craig

Craig

    Vendor (Stark Labs)

  • *****
  • Vendors
  • Posts: 768
  • Joined: 16 Sep 2003
  • Loc: Irvine, CA

Posted 05 July 2012 - 05:34 PM

UPDATE --

First, if you can't see the graphs well, I apologize. My PNGs got converted to JPEGs and in the process got both larger and fuzzier. Go figure... I have a PDF up here: http://www.stark-lab...onLinearity.pdf

Second, I've spent the lunch break (and then some) taking some shots to look at the ISO vs. dynamic range issue in some more detail, motivated in no small part by a nice e-mail from Jerry Lodriguss. The test here, rather than plotting the mean intensity value in a flat of varying brightness, shoots a target at very low light using a solar filter. It's a perfectly overcast day here which meant that I could shoot outdoors reliably and not worry about intensity changes.

Back in my 2009 review, I pegged the optimal ISO for the cam at 800 and if you go through the full-well vs. read noise and if you look at how the read noise drops until ~1000 ISO and then flatlines with no benefit in going higher (with most of the win done by ISO 800), you could see my motivation.

Here, my data was showing better apparent low-light performance in ISO 400 than ISO 1600. Yet, as Jerry pointed out, if you calculate the dynamic range by max-signal divided by read-noise, the ISO 1600 should be better. You've given up 2 stops in terms of the top end but the dynamic range is only down 1 stop, meaning you must have made it up 1 stop somewhere -- in this case on the low end. My data average over may pixels (and we average over some number of frames) and that's part of it. But, the more I thought about it, the more that last figure seemed to not add up. The top two panels should be curvilinear and the bottom ones be linear. ISO 400 doesn't follow that as well as ISO 1600.

Now, no matter what, each time you up the ISO beyond 200, you're cutting out signal on the top end. The question is what's happening on the bottom end. You're losing dynamic range - the question is how much. At least from looking at individual frames (and not stacks), on this cam it looks like 800 may be the best bet for optimizing the low-end. There is a bump up from 400 to 800 in terms of how much you can pick up. 800 and 1600 is splitting hairs on the low end. Here, keep in mind, that we're letting read noise dominate and in a number of situations, it'll be swamped by other sources of noise. Your total dynamic range is also being cut given the harsh chops off the top.

None of this, though, suggests that you should be going to 3200, 6400, or 12800, etc. Boosting the ISO helps to a point, but only to a point. After that, it's cutting the top end and not giving you anything in return on the low end. That conclusion I'll stick to even if we bump it up a stop.

Craig

#9 Craig

Craig

    Vendor (Stark Labs)

  • *****
  • Vendors
  • Posts: 768
  • Joined: 16 Sep 2003
  • Loc: Irvine, CA

Posted 05 July 2012 - 05:36 PM

Here is my question. Do you think that the "correction" of the image is done in firmware or software? Would simply programming some new software for the camera fix it? Have you tried the green lantern software?


It's in the CR2 file as the camera writes it. I don't know green lantern, but I do know Canon's lossless-JPEG compressed TIFF-esque format. Both dcraw with those parameters and Nebulosity just do the simple decompression. Canon's own software, along with others I've tried that are geared towards daytime work do a lot more to it from there.

Craig

#10 Chris Todd

Chris Todd

    Lift Off

  • *****
  • Posts: 6
  • Joined: 17 Aug 2008
  • Loc: Catonsville, MD

Posted 09 July 2012 - 11:19 AM

Craig,
Forgive me if this is a stupid question, but I know there are two menu options buried deep within most recent Canon camera menus - High ISO noise reduction, and long-exposure noise reduction. Did you perform your tests with those options enabled or disabled?

Clear skies,
Chris Todd

#11 Craig

Craig

    Vendor (Stark Labs)

  • *****
  • Vendors
  • Posts: 768
  • Joined: 16 Sep 2003
  • Loc: Irvine, CA

Posted 11 July 2012 - 08:57 AM

All inboard camera processing / noise reduction has been disabled in the menus.

Craig

#12 Wyatt Earp

Wyatt Earp

    Lift Off

  • -----
  • Posts: 7
  • Joined: 25 Jun 2008
  • Loc: Canada

Posted 25 March 2013 - 08:58 AM

Hi Craig and thank you for the review. I was considering picking up a XSi for astro and this helped answer some of my questions and more.

Additionally, would you recommend a current offering from Canon vs a used XSi?

Cheers,
Wyatt.

#13 timmbottoni

timmbottoni

    Surveyor 1

  • -----
  • Posts: 1756
  • Joined: 25 Aug 2005
  • Loc: W Chicago suburbs, IL USA

Posted 19 July 2013 - 12:52 PM

Any guess what the optimum ISO for a Canon 60D would be?

Thanks

Timm

#14 derangedhermit

derangedhermit

    Apollo

  • -----
  • Posts: 1158
  • Joined: 07 Oct 2009
  • Loc: USA

Posted 20 July 2013 - 10:33 PM

The 60D, according to the daylight tests at DXOMark, has maybe a half-stop DR improvement over the xsi, a slight improvement in noise across the board - you can draw your own conclusions by looking up each camera there at the sensor tests. The problem is that the results don't tell us much about long-exposure performance. My *guess* is that "ISO 800" is still about as good as any - the greatly increased ISO range availability is not matched by high sensor performance in the extended range.

I'd like to rephrase the question to be

"How do we find out for our camera what the optimum ISO is (without going to all the trouble you did)?"

lee






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics