Confusion over high ISO, noise, and light levels

Looks interesting. Even on the full sized image the one on the left looks poor in comparison. Think I might be out playing around with my 50mm tomorrow :) I've not actually noticed this phenomenon on my sony at all. In the low light workshop I did I was using max aperture and iso 1600 all the time and the noise levels seemed to be perfectly ok and the same as usual for that iso speed.

I wonder if some camera makers actually slightly increase the exposure time to compensate instead? It would be interesting to time the shutter speeds and see whether there was more deviation at wider apertures.
 
Yes, about one stop vignetting on a 7D tallies with the-digital-picture link.

I see what you're getting at, and it's well known that sensors are sensitive to the angles at which light strikes them, but I also think your test image is very deceptive.

It needs to be an area of plain tone, possibly deliberately completely out of focus to eliminate the optical shortcomings at f/1.4?

OK, I re-shot with extremely OOF subjects. This was indoors under tungsten and then WB corrected. No other edits. Here is f/1.4 vs f/2.0 because even at f/2.8 there were colour patches emerging in the distant background, making direct comparison more tricky. These are centre frame crops at 100%.

20110818_195651_000.jpg


It could be my imagination, but I think the shot at f/1.4 might look a bit grainier - courser - especially at the transition between the lighter and darker patches at the bottom of the crop. I think we've been there before. :D.


EDIT : oh, and the context again....

20110818_200535_000.jpg


I didn't notice earlier, because I was looking for noise, but it seems evident here that the shot at f/2.0 is brighter than the shot at f/1.4, for what appears on the surface to be equivalent exposures. Sampling equivalent areas in Lightroom bears this out. Hmmmmmm! :thinking:
 
Last edited:
As you say, hmmm... It's not really anything to lose sleep over.

No question it's darker at f/1.4. And if you look at the darker tone, and not the light/dark transition which is skewed a bit by bokeh fringing, it's so close that equalising the exposures correctly would effectively level the noise?

As to why the two are different exposures, it could be that the lens is not quite f/1.4 and 1.45 or something, and that at f/2 it's actually f/1.9. Quite possible. I think shutter speed is likely to be pretty accurate.

Or it could be that, in tweaking the ISO to sort out the sensor issues, Canon hasn't got it quite right... LOL

Hope your sore throat gets better ;)
 
This is such an interesting thread I think it deserves a bump so that others who missed it first time round (like me!) can enjoy it too.

Tim, is your throat ok now? ;)
 
Seven months on I'm just about over the worst. Thanks for asking. :D
 
Also, digital sensors do not suffer from reciprocity failure as film does.

Great thread revival this :thumbs:
I missed it first time around.
Interesting stuff; and yes, my brain is melting.

I'm not sure about Hoppy's statement.....
I'm convinced my 20D suffered from a digital equivalent.
I always had to dial in a little extra exposure beyond the meter reading for shutter speed over 6s or so, and a lot extra by the time the duration got to 30s.

By comparison, the 5DII exposure metering is spot on even at High ISO & 30s.
Maybe the 20D just has particularly crap metering :shrug:
 
FWIW I have applied no adjustment in Lightroom. These are unedited raw comparison. Hopefully Lightroom is not applying different processing to the two raw files, including vignetting correction.

Excuse me if im not getting your point BTW.

Except the image algorithm has been applied. ..which is what does the damage it seems... at leasts that where we got to I think. :)

I see the problem as us looking to hard at kind of early technology and comparing it to the god of film. I see it as being down to the art of the hardware and software engineers that goes much of the way in the final consistency of images produced, errors are corrected, magic is performed and the end result is a compromise for best overall effect ..as technology gets better, the bar is raised again.
 
Excuse me if im not getting your point BTW.

Except the image algorithm has been applied. ..which is what does the damage it seems... at leasts that where we got to I think. :)

I see the problem as us looking to hard at kind of early technology and comparing it to the god of film. I see it as being down to the art of the hardware and software engineers that goes much of the way in the final consistency of images produced, errors are corrected, magic is performed and the end result is a compromise for best overall effect ..as technology gets better, the bar is raised again.

I think this thread is simpler than that and if one has a better understanding of the possible causes of the issue one has a better chance, potentially, of doing something to avoid or reduce it. The thread has nothing to do with film or artistry or technical advancement, as far as I can see. The nub of the question is....

"Why does a shot taken at f/8 in good light look cleaner than the same shot taken at f/1.4 in lighting 5 stops dimmer with no adjustment to shutter speed or ISO?"

On the face of it you'd think you would be capturing the same number of photons and basically have the same amount of noise. But, in the thread we have put forward some ideas which may or may not have merit, but seem to be plausible. They are, in no particular order....

1. When you shoot with shallow DOF (wide apertures) noise can appear in areas where previously it was not apparent. This includes smooth, featureless OOF areas and also transitions at high contrast edges.

2. If light levels do vary by 5 stops it is possible/probable that the colour temperature becomes skewed and adjusting white balance to compensate may well increase noise in the channel which was starved of light. Tungsten light (very orangey red) will cause blue channel noise. Shaded daylight (very blue) will cause red channel noise.

3. Fast apertures may gather more light than slow apertures, but the efficiency of light gathering reduces the wider/faster you go. This means that if you maintain the exposure level by the numbers alone you may end up underexposing more as you open up the aperture

4. With very fast lenses and wide apertures light falloff in the corners is likely to be worse than when stopping down. This is closely related to point 3, but emphasised at the edges and corners. If you choose to correct that in software then you could well find yourself needing a 2 stop push in the corners, which inevitably will have a negative impact on noise in those areas.




As for comparing files, be they raw or JPEG, one way or another they will have been processed by software to turn them into a visible and usable image. If you've ever seen a raw file in its native state you'll see it's a flat looking file with a heavy green tint. Here's an unprocessed raw image, which was shot with a good exposure and decent white balance, but the bare raw file looks very odd indeed. In this image the house is painted white and the sky in the background is featureless grey cloud. The green tint is no surprise since the Bayer array has twice as many green pixels as red and blue.

20120315_170636_000.jpg



It needs to be software manipulated to be useful as a picture. The question we can't really answer is whether there are any "under the covers" adjustments made by the rendering software before we get our hands on it. For example, if you enable HTP (Canon feature) your raw file will be underexposed by 1 stop courtesy of a secretly reduced ISO value. DPP will see that HTP was enabled and apply a special tone curve to adjust how the image is displayed. I don't know how other software tackles the issue. It was the case, and possibly still is, that Lightroom would blindly double all the raw data values throughout the tonal range, just emulating a brighter exposure in the first place. You also have picture styles and camera profiles to consider. Whichever way you look at it the files have to be processed and do we really know if there are any little secret tweaks going on? I don't. I have to work on the assumption that the playing field is level, but who knows?


So, at the end of all that, nothing to do with film, nothing to do with art, nothing to do with technology advances, just some ideas to be considered when trying to reduce noise in these low level lighting scenarios. Is fast glass always the best answer (where noise is concerned) or could we get cleaner files by stopping down a bit and reducing shutter speed, if that option is viable? Is natural light always the best answer or would we be better off chucking some flash into the mix? Do we shoot tight and fix the vignette, or shoot loose and crop it off? Or do we say "Screw the noise, I'm an artist." and I'll shoot however I damn well please? :D

Better to know about these things and weigh up the options than to know nothing at all.

p.s. There's always Neat Image :thumbs:
 
I think this thread is simpler than that and if one has a better understanding of the possible causes of the issue one has a better chance, potentially, of doing something to avoid or reduce it. The thread has nothing to do with film or artistry or technical advancement, as far as I can see. The nub of the question is....

"Why does a shot taken at f/8 in good light look cleaner than the same shot taken at f/1.4 in lighting 5 stops dimmer with no adjustment to shutter speed or ISO?"

<snip>

In practise, there are so many other things going on when the light shifts five stops that it's impossible to draw too many firm conclusions without doing a controlled test.

Here's one change that often seems to get overlooked, and would certainly apply in Tim's example - how lens MTF performance is affected by aperture, and what it actually means.

Folks usually think about sharpness in terms of resolution (the fineness of detail) when it's actually contast (how clearly those details are shown) that gives us the visual impression of 'sharp'. What matters much more is % contrast at say 30-40 lpmm (ie lens MTF performance) than micro detail at 100 lpmm that you'll struggle to even see.

Example - this is what happens when a Leica 50mm f/1.4 lens is used at maximum aperture and then at f/5.6. Go to this link, then click on Technical Data on the right http://en.leica-camera.com/photography/m_system/lenses/2183.html

It will load a pdf with three MTF graphs. Look at the top two, at 1.4 and 5.6, and note for example, how the bottom set of lines (at 40 lpmm) jump up on the left from about 57% at f/1.4, to 70% at f/5.6. That is a very noticeable improvement in visual sharpness, and it will go a smidge higher at f/8.
 
Last edited:
So, at the end of all that, nothing to do with film, nothing to do with art, nothing to do with technology advances, just some ideas to be considered when trying to reduce noise in these low level lighting scenarios. Is fast glass always the best answer (where noise is concerned) or could we get cleaner files by stopping down a bit and reducing shutter speed, if that option is viable? Is natural light always the best answer or would we be better off chucking some flash into the mix? Do we shoot tight and fix the vignette, or shoot loose and crop it off? Or do we say "Screw the noise, I'm an artist." and I'll shoot however I damn well please? :D

Better to know about these things and weigh up the options than to know nothing at all.

I think you know my answer to what I'd say, the image comes first... then good exposure of it ...and sharpness (their is a thread about sharpness not being important ;) ) ...then everything else. I care, just not that much.

Thanks for making that clear to me Tim :thumbs:... got it now, yes interesting observations, luckily, we can, if we choose, leave all these usefull ideas to the manufactures to deal with, I've little doubt that they aren't aware of all these problems and far more, and Id go some way to bet they have allready given us the best comporomise with the technology available.

Even if my old D70 is getting to be a bit poo in comparison nowadays ... ;)
 
Last edited:
Back
Top