Long Exposure Sensor Testing

[ruifo]

This is an interesting analysis of long exposure for several cameras:
Long Exposure Sensor Testing. | Brendan Davey Photography

This is a useful test for astrophotography.

The D810 was missing, so I just tried to reproduce this test (as of the source) with my D810, but using a different software to convert (Lightroom 5.6), and converting from RAW to JPG, not to DNG. So there could be differences. All other paramters are the same of the source. Anyhow, here are my tests, both in full RAW, as well as in sRAW, as the D810 has both formats.

 

[Josh66]

What did it look like before you bumped the exposure by 2 stops?

 

[ruifo]

What did it look like before you bumped the exposure by 2 stops?

Here, only changing the exposure back to 0 (zero):

 

[spacediver]

although not directly related, you might find this interesting. The third and fourth image are some preliminary noise measurements. In the near future I'll be doing more extensive noise testing (although only at ISO 100).

 

[D-B-J]

Wow, the D3s and 6D are horrid!

 

[ruifo]

Other FX cameras are missing, like the D600, D610, D700, D4S. Could someone who owns one of them try to do this test with them?

 

[astroNikon]

one reason I try to keep all astro stuff to under 30 seconds. The crop cameras require longer exposures for similar results, which created even worse black results. I have to try my hand at stacking too one of these days.

I can try redoing this test on my d600 & d7000. I did them a while ago.
these base numbers may interest you too ... I'm putting together a more complete chart including Canon .. the chart is below the DX cameras ==> dxoMark Nikon ISO numbers | Photography Forum

 

[astroNikon]

check your inbox for info ...

 

[ruifo]

With astroNikon's help, here is the results of this test for the Nikon D600 as well:

________________________________________________________


Here, only changing the exposure back to 0 (zero):

 

[jsecordphoto]

Wow, the D3s and 6D are horrid!

Yup. I do lots of long (5 minute+) exposures with my 6d for night photos, and the noise is pretty gnarly. Long exposure NR and removing color noise in LR helps a lot, but still. Glad to see somebody is doing these tests

 

[Stradawhovious]

Wow, just with preliminary testing using the settings from that site, my D7000 is a rockstar.

Unless I'm doing it wrong.

 

[BrendanDavey]

Hi Guys,

I have now followed the same workflow and uploaded the results for the 610 and 810. If I was going to buy Nikon at the moment I'd be buying the D750 The newer 810 and 610 outperform the old 800 but I think NIkon are right when they said the newer sensor in the D750 is better. As mentioned on the page I suspect that the test Dx0 and dpreview etc perform are very short exposures in comparison. If I was going to buy a Canon, the 5Dm3. The 6D holds up well for shorter periods and beats the 5D, but at 5m I think the 5D has less noise.

If people are interested in crop bodes I can look into adding them as well...

I'm hoping to add the A7s and 1Dx soon.

Cheers, B

 

[TCampbell]

Physical camera temperature (how hot was it outside and how long had the camera been in use before doing the test) will impact these results as well. You can MASSIVELY decrease noise by physically lowering the temperature of the camera sensor. In frigid cold nights here in the north, it's much less of a problem.

However, the image registration and integration software can do wonders for this if you tell the auto-guider to dither when you're capturing your "light" frames.

The idea is that although the guider is keeping the scope tracking during image acquisition, you'll get both random noise and pattern noise in the frame in addition to the real data (e.g. stars, etc.) and at some point you wonder when something is "noise" vs. a very very faint background star. So you take multiple "light" images and "stack" them together using image processing software (I use PixInsight).

The noise can be knocked back by a factor of the square root of the number of frames you took. So if you take 9 "light" frames, you can reduce the noise by a factor of 3. Take 16 and you can reduce the noise by a factor of 4. This is because as long as the mount and scope are solid and tracking nicely then everything representing "real" data will appear in every frame and will occupy the same pixels, but anything appearing randomly (e.g. noise) will land on random pixels and the computer should be able to determine (via the stacking algorithm and you get to control which algorithms are used.) when it's "noise" vs. "data."

Sounds good, right? Well not quite.

Turns out you also get "pattern noise" where the same pixels tend to get lit up in each frame (given the same conditions when the image was taken and that includes physical temperature). That means the computer will think the pattern noise is real data.

But this is where the "dithering" comes in.

If you are using guiding software that integrates with your image acquisition software (e.g. I use PHD guiding and BackYard EOS for camera control and image frame acquisition when I'm at the observatory) then once the shutter closes, the image acquisition software will tell the guiding software to randomly move the scope by a few pixels. Once the auto-guider finishes moving the scope (it doesn't move far) it tells the image acquisition software that it's ready for the next "light" frame to be captured. You end up with a stack of "lights" which are each randomly offset just a tiny bit. But NOW your "data" will appear to be slightly shifted (each star will have moved a tiny bit) in each frame... meanwhile the "pattern" noise will not have moved. The "registration" component of the image acquisition software aligns each frames based on the stars and this causes the "pattern noise" to now appear in somewhat random spots. The result is that the computer can now, finally, figure out that this is, in fact, just "noise" and can safely be removed from the final integrated image. It's a beautiful thing when it all comes together like that.

BTW, for this to work it's better to use a stacking algorithm like sigma-clipping vs, say, "averaging". When you "average" frames the bad data just gets dimmer but isn't completely eliminated. But sigma clipping works a bit more like a winner-take-all "vote" for each pixel's value. E.g. suppose you've shot 16 "lights" of an object, but in just ONE of them... a satellite or aircraft went flying through the frame (something that happens more often then we'd like.) With 'averaging' that streak from the satellite or aircraft will get dim. But with sigma-clipping it's a bit different. The computer checks to see what color that same pixel was in frame 1 (black) vs. frame 2 (black) vs. frame (3) black, and so on... but then let's say in frame 16 it was lit up brilliantly due to the aircraft lights. Since the computer effectively has 15 frames that "vote for black" and 1 frame that "votes for white" then black wins. The computer discards the "white" value from that one frame that disagrees with the rest ... it's an outlier... and the result is a nice image. Averaging works best when you just have a few frames. Sigma clipping is better when you have lots of frames.

As for Back Yard EOS... that uses the Canon SDK to control the camera so it won't work on a Nikon camera but I hear they're developing a Back Yard NIK. I haven't heard anything on that in quite some time so I have no idea how it's progressing.

 

[ruifo]

Hi Guys,

I have now followed the same workflow and uploaded the results for the 610 and 810. If I was going to buy Nikon at the moment I'd be buying the D750 The newer 810 and 610 outperform the old 800 but I think NIkon are right when they said the newer sensor in the D750 is better. As mentioned on the page I suspect that the test Dx0 and dpreview etc perform are very short exposures in comparison. If I was going to buy a Canon, the 5Dm3. The 6D holds up well for shorter periods and beats the 5D, but at 5m I think the 5D has less noise.

If people are interested in crop bodes I can look into adding them as well...

I'm hoping to add the A7s and 1Dx soon.

Cheers, B

Very good!!

 

[astroNikon]

Wow, just with preliminary testing using the settings from that site, my D7000 is a rockstar.

Unless I'm doing it wrong.

The d7000 supposedly is really good too. One reason I originally went with that model. Alot of astrophotographers loved the d7000 for it's blackless something or another.