Overexposing ISO?

[r_aysh]

i STILL DONT GET HOW IT WORKS,

It happens to me, I get different noise results even at the same ISO with same

light situations..its just that there would be a change in exposure measures..

Now i have been lucky with guessing the best exposure yet I need to know what

Im doing exactly..

Help Please!

-Im have the noisy D80

[tonybeach]

"I get different noise results even at the same ISO with same light situations..its just that there would be a change in exposure measures..."

 

Noise is a result of underexposure. Raising ISO does not change exposure, it raises the amplification of the digital data and is analogous to pushing film stock. If you have two different exposures of the same scene taken at ISO 400 and one is underexposed by one stop, the underexposed shot will have the equivalent noise to a "properly exposed" shot taken at ISO 800.

[geoffs1]

Anthony Beach"...two different exposures of the same scene taken at ISO 400 and one is underexposed by one stop, the underexposed shot will have the equivalent noise to a "properly exposed" shot taken at ISO 800".

 

If it were only that simple. A properly exposed ISO 800 shot should have better signal/noise than an ISO 400 shot underexposed by 1 stop (i.e. taken with the same shutter/aperture combination as the ISO 800 shot).

[alvin_wong]

<p>Ryan, I suggest you read the excellent tutorials on exposure on this website. If you want consistently better pictures, you have to stop guessing and start exposing properly for the picture that you want.</p><P>Still, keep in mind digital SLRs are a lot like slide film in that highlights blow out easily. At the same time, you can't underexpose too much, unless you want noise. You need to decide what is more important to you: shadow detail, or highlights. Sometimes there is enough exposure latitude for both, but usually you have to sacrifice one for the other.</p>

[alvin_wong]

<p>I just wanted to clarify what I just said. In scenes with moderate contrast and limited dynamic range, you can generally squeeze in most, if not all of the dynamic range. Still, proper exposure is generally needed, if you want clean, noise-free pictures.</p>

[tonybeach]

For Geoff:

 

"If it were only that simple. A properly exposed ISO 800 shot should have better signal/noise than an ISO 400 shot underexposed by 1 stop."

 

It always amuses me when people say "should" in their rebuttals; evidence of what actually happens is better than opinions and conjecture. I have done the tests and it is clear that what is happening to the RAW data in Nikon's DSLRs (I haven't done this testing on the D300, but it is irrefutable for all of Nikon's earlier DSLRs), the noise levels are exactly identical between ISO 400 pushed one stop and ISO 800 with the same aperture and shutter speed settings. NR and contrast applied by the camera will change the outcome though, so you may think your ISO 800 shot has less noise than your one stop underexposed ISO 400 shot, and it will if you are shooting JPEG and not RAW and disabling auto contrast and NR during conversion. If you have evidence to contradict my experience, then by all means share it here.

[geoffs1]

Anthony: you're quite right that I haven't actually done any measurements (strange, because that's the exactly the kind of thing I do for a living - perhaps it's the "busman's holiday" effect). However, the noise caused photon-counting statistics, the CCD, and the analog amplifier gain are usually not so straightforward as to combine as you describe. That's not to say it could happen under certain circumstances. So, I don't doubt your measurements (esp. if they were done using raw files).

 

In the situation you describe, why do you ever bother changing the ISO setting?

 

I think some of the astronomical photographers have characterized the noise sources in some modern digital SLRs. I'll have to see if I can dig up the reports.

[geoffs1]

<p><a href="http://www.clarkvision.com/imagedetail/digital.signal.to.noise/">Here's the article</a> I was thinking of.</p>

[frank_skomial]

Never overexpose, but instead use higher ISO and expose properly, for too many reasons to count...

[geoffs1]

"Never overexpose" - did you mean "underexpose"?

 

I agree you want to avoid overexposing, but in the context of the rest of you sentence "underexpose" would seem to make more sense.

[tonybeach]

For Geoff:

 

"In the situation you describe, why do you ever bother changing the ISO setting?"

 

It has a lot to do with which conversion software is being used. I generally prefer Capture One (by Phase One, not be confused with Capture NX or Nikon Capture) and that program handles files differently based on their ISO settings (appears to be some behind the scene and unadjustable NR). Also, the most you can push a file using Nikon's software is two stops before having to resort to measures that will distort the relationship between EVs within the file.

 

What is important to understand is that with Nikon's DSLRs using higher than base ISO is underexposing. In fact, the greatest irony about advice routinely dispensed on the internet about how to expose with Nikon's DSLRs is that people will be advised to underexpose because they are mistakenly being told to determine exposure based on the camera's red channel histogram and then think they are correctly exposing by turning up the ISO. It is little wonder that with bad advice being compounded that many decide that the camera is deficient and clamor for the next "better" model that will solve their problems.

 

For Frank:

 

I am patient and willing to read all of your reasons. One reason to not to boost ISO unnecessarily is that there is more DR at lower ISOs; I often see highlights being blown by photographers so that shadow detail that could be recovered just as effectively during conversion and PP is "properly exposed". The best advice about exposure I can give is to use the lowest available ISO (although not necessarily ISO 100 on the D300 or D3, they should only be used when necessary) and to ETTR within reason (not all highlights have to be preserved in all channels, you can burn highlights in the individual channels to get blue back into blue skies for instance).<div></div>

[rene gm]

I did a short test (I shouldn't not to contribute to all that noise fuss) with three times underexposure, and found no difference worth mentioning (RAW and ACR). I'd be surprised otherwise, since the difference is only the place where the amplification takes place. Higher ISO pushes in camera, and underexposing in software.

 

This is for RAW. Note that pushing the camera generated JPG is different, since shadow details may have already been nullified.

[marco_p1]

Anthony, it would be very interesting to see a couple of real world images with the same processing you show in your gray patches. I will try myself, but maybe you have sometfing already at hand...

Thank you,

Marco

[ilkka_nissila]

Normally the ISO setting would affect analog amplification in the camera and this generally producess less noise than underexposing in the camera at a lower ISO and increasing the brightness in post-processing. You might need to do your tests with a difference of more than 1 stop to see the difference.

 

If you underexpose at base ISO and then compensate in post-processing you will lose more dynamic range in the final picture than you would if you had used an appropriate ISO and exposure in the first place.

[tonybeach]

"Normally the ISO setting would affect analog amplification in the camera..."

 

It actually happens in two stages on Nikon DSLRs. The analog (on chip) amplification occurs in 1/6 steps from base ISO to ISO 360. After that, the amplification is performed by a signal gain amplifier, so then it is two amplifiers working together.

 

ISO 400 = analog amplifier 2x + gain amplifier low (2x)

 

ISO 500 = analog amplifier 2.5x + gain amplifier low

 

ISO 800 = analog amplifier 2x + gain amplifier high (4x)

 

ISO 1600 = analog amplifier 2x + gain amplifier uncalibrated (8x)

 

Note that amplification is of ISO values and not stops. Nikon seems to have fine tuned the second amplifier between the D200 and the D80:

 

http://www.dpreview.com/reviews/nikond200/page22.asp

 

http://www.dpreview.com/reviews/nikond80/page19.asp

 

The graphs suggest that high gain amplification on the D200 takes a 1+ stop hit in DR, while the D80 holds up better until it reaches uncalibrated gain amplification.

 

"You might need to do your tests with a difference of more than 1 stop to see the difference."

 

The test image presented here is 2 stops. Here was one I did a while back at 3 stops: http://photos.imageevent.com/tonybeach/mypicturesfolder/rawconvertors//100-800%20Comparisons.jpg

 

"If you underexpose at base ISO and then compensate in post-processing you will lose more dynamic range in the final picture than you would if you had used an appropriate ISO and exposure in the first place."

 

Since DR is no different than the distance between highlights and acceptable noise, this is not correct. What you lose is bit depth in file (mostly in the highlights) as the data is being captured at a lower bit depth and is captured linearly on Nikon's DSLRs. Since no one can see the difference between a compressed NEF (essentially 10 bits, with most of the lost data in the highlights) and an uncompressed NEF (full 12 bits; something the D80 ironically does not have), the value of that bit depth is dubious: http://forums.dpreview.com/forums/read.asp?forum=1021&message=26119816 (this is a worthwhile discussion started 3 days ago about this very topic and the poster I link to is one of the authors of Raw Magick Lite, the RAW converter that I consider to be the best in terms of image quality).

 

Much of this discussion has become highly technical and I suspect of little use to Ryan, but the point is that underexposure causes noise in the shadows. We always say to expose to the right (ETTR), but it would be more accurate to say expose away from the left. However, just because your histogram says your data is not underexposed; in reality it is and changing ISO is amplifying it. Why amplify an underexposed shot in the camera, even if it is the RAW file? If you are using ACR or Capture One (to name two that I am aware of, I'm sure there are others but Raw Magick Lite and Nikon's software adjust EC linearly), you are better off doing as little EC as possible because ACR adjusts exposure non-linearly and Capture One appears to apply NR based on the selected ISO; and in general workflow is facilitated by files that require minimal adjustments prior to conversion.

[ilkka_nissila]

Above in your post ofDec 20, 2007; 05:52 p.m., your illustration shows and image correctly exposed at iso 800, and a second one at iso 200 underexposed by 2 stops (i.e. given the same exposure as the iso 800 image) and corrected to match the iso 800 image with +2 exposure compensation in post processing? If this is the case, you have more noise in the post-compensated iso 200 image according to the numbers. I did a similar test using iso 100 (underexposed by two stops) and then corrected in Nikon capture by applying +2 EC. This image was compared with an iso 400 image and there was more noise in the shadow areas of the underexposed and post-compensated iso 100 image than in the iso 400 image which was correctly exposed. More noise in the shadows => reduced dynamic range. Thus the results from underexposure with post-processing exposure compensation are worse than those results from using the correct higher ISO in the first place in terms of noise and dynamic range.

 

Thus, as expected, the best results are obtained using the correct ISO and exposure in the first place.

[tonybeach]

"Thus, as expected, the best results are obtained using the correct ISO and exposure in the first place."

 

I posted the standard deviation differences clearly on the image. Consistent with what can be seen visually in my example, there is so little difference in noise that its effects are essentially irrelevant (perhaps 1/10 of a stop or less). What you are insisting on saying is that there is a different exposure value between two shots with the same aperture and shutter speed but with different ISOs; that is wrong, any exposure at more than base ISO is underexposed (even if the highlights are blown!).

 

My point in this thread as it relates to the OP is simple. If you underexpose an ISO 100 shot, you will have the same noise that you would have had if you had taken the shot at a higher ISO (or negligibly more if you are incurably fixated on unperceivable minutia).

 

Marco has asked me to show a "real world" example in this thread, but that is not very scientific. Instead, I will now post the full frame shots (downsized so they are viewable in thread). This is an experiment anyone can do, and done correctly the results will show standard deviations of less than .4% (my results with Raw Magick Lite were about .2%, half as much an with NX) in any channel in any color patch. These shots were taken in low diffused daylight inside my home within seconds of each other.<div></div>

[tonybeach]

Here's the other shot I used in my first example (although that one was originally processed with NX).<div></div>

[ilkka_nissila]

Anthony, the difference in standard deviations you show is not negligible, it is a significant difference.

 

I didn't say that there was a different exposure value. The exposure is the same but for one ISO setting it is the correct exposure (in your case, iso 800) but for the other (iso 200) the shot (using the same exposure value) you get a file which is exposed two stops under. This file is more noisy when you adjust the image to compensate for the underexposure than the correctly exposed iso 800 image (as your standard deviations show).

 

I compared the shadow noise on 1) an iso 100 image exposed according to iso 100 to 2) an iso 400 image exposed at iso 400, to 3) an iso 100 image exposed at iso 400, and the 3rd had 30% greater noise increase from the first than the second. So this 30% greater increase in noise is why you should use the iso dial instead of just underexposing.

[tonybeach]

"So this 30% greater increase in noise is why you should use the iso dial instead of just underexposing."

 

I am not sure what you mean by 30% more noise or what converter you are using or what settings are being applied (sharpening, NR, contrast, etc.).

 

Anyway, here are my numbers from the darkest patch (brighter patches had similar or smaller numbers, which means the standard deviation percentages were over five times less in the brightest patches). Begin by understanding that the average RGB value for this patch is 48 (one exposure read 49 and the other 48, a difference in exposure value of 2%). Below is the difference between the standard deviation for each channel (and yes, the ISO 200 +2 EC always did slightly worse, but just how slight will soon become evident).

 

Red channel is: 4.92 v. 4.66 = .26 (.5% more noise)

 

Green channel is: 3.99 v. 3.63 = .36 (.7% more noise)

 

Blue channel is: 69.02 v. 5.67 = .34 (.7% more noise)<div></div>

[tonybeach]

Correction: first column of the blue channel should have been 6.02, not 69.02.