Why increase ISO digitally with RAW mode

[Rainer SLP]

This is interesting because I had a situation where I needed to use iso 800 to make a photography of a cave.

If I would not have switched to iso 800 the speed of iso 50 or 100 would have been impossible to handheld and I did not have a tripod with me.

So if I shoot everything with iso 80 as Kodak wants us to do, I have to have a tripod with me everytime and everywhere. Right?

Or did I understand your question wrong?

[Rainer SLP]

Inspired in your post I made 2 shots.

1st iso 400 1" f16
2nd iso 100 1" = -2 f16

Then I worked them both in Photoshop 7.0 to get the same from both.

I liked the original iso400 more than the worked iso100.

I prefer to adjust the camera to iso400 when I need it rather than at site having to begin to think which speed and what compensation do I have to adjust to get the same speed, etc. etc. etc.

I think the camera producer guys (at least Canon, because that is what I have a 1Ds) do have made their homework in developing the right digital pushing of the sensor directly in the camera.

[Rainer SLP]

In iso100 -2 compensated the image was noisier than at original 400.

On the ohter side I printed yesterday the iso400 cave shot as a 12 x 18" print and I was astonished as how low the grain is. OK I had to photoshop it, but my scanned slides had to be photoshopped too, so digital 400 is far better than analog 400.

Well there were no highlights. I will try again with hightlights.

On my last photo trip I took a look at the histogramm and the highlight warning and in most cases -1 compensation was OK. I mostly had this problem in landscape with sky in it or very bright rocks. OK it was between 11:00 and 14:00 hours. In very difficult situations I think you will have to live with a few highlights.

I will try it right now and post some sample inages so you see them and can make up your mind.

Added half an hour later. While shooting it came to my mind that you also have the possibility of choosing you WB mode and here as far as I have seen also some highlights are already compensated.

So all this is not easy, it gets more complicated from shot to shot.

Anyhow it also interest me, because making this Bracketing shooting you learn for real life situations  

[willie408]

You are probably gonna remain puzzled, too.  I am quite sure the software in a digital camera is a very closely guarded and important secret.  I will be you anything that when new "firmware" (what an odd name for software) is downloaded into the camera it is encrypted and that decryption takes place due to the existence of the previous version of firmware.

[BJL]

I read recently comments from Kodak suggesting that digital sensors have a quite low natural ISO (50-80?) and most or all of the ISO adjustment on digital cameras is done digitally, after A/D conversion, with little or none being amplification in the analog stage.

Digital amplification makes sense if the data is then going to be cut down to 8 bits, because it throws aways useless "leading zero bits" making room for more of the somewhat useful less significant bits. But it seems to make no sense if the file is going to be store RAW (or 16 bit TIFF): at worst, it risks blowing out highlights that the camera had recorded successfully if you overstimate the ISO needed.

Why not instead offer only a smaller range of ISO adjustments through gain settings for the analog amplifier section, and leave the rest to "digital push processing" in the RAW conversion or editing software?

Do some DSLR's offer that more conservative option?

Can one find out which is the lowest ISO setting for each amplifier gain level and use only those, to avoid "in camera digital pushing"?


P. S. [added] If I recall correctly, the Kodak 14n uses more that the usual amount of variation in analog amplification levels as one varies the ISO setting, which might contribute to its greater variation of noise level with changing ISO [Did I phrase that diplomatically enough?]

[BJL]

*IF* I am right about most or all of the ISO increase being "digital pushing", then one would work as if one were going to push process film: set the aperture and shutter speed as if you were using the higher ISO, getting an initially underexposed RAW file to avoid possible highlight problems. In your cave example, in place of using ISO 800 "on meter", you might use ISO 100 underexposed by three stops to get the same shutter speed and aperture.

In situations where there is no risk of blowing out highlights, there is probably no need to bother, and in camera ISO pushing is the most convenient way to go, but with a high contrast scene (a shaft of sunlight in your cave?), my proposal might be the cautious way to proceed.


Disclaimer: I mean it about "IF I am right"; I am still hoping for expert technical feedback.

[BJL]

Thanks for the experiment! (My far humbler digicam is too poor at ISO200+ to bother; I ask these questions in my search for the best next digital camera to get).

What is the difference that you see between "ISO400 straight" and "ISO100 at -2 stops with brightness turned up in Photoshop"? How does the noise in shadow areas compare? Is the contrast changed? Was it a very high contrast subject? (You have already persuaded me that only high contrast subjects could possibly benefit from the extra effort I am talking about.)

And does anyone know if the 1Ds changes the analog amplifier gain between ISO100 and ISO400, or only the digital gain? Such a change is my first guess at the source of any differences that RS Foto has seen.

(I suppose that a variant of what I propose for avoiding highlight blowouts would be metering on highlights, placing them at what one knows is a safe "zone", say +3 or +4, and then choosing the lowest ISO speed setting that gives a safe shutter speed.)

[samirkharusi]

I first acquired a consumer DSLR when the D30 came out and this concept of varying ISO has always bothered me. If anyone is interested in high ISO it must be astronomers, and astro CCDs have NO setting for adjusting ISO. You do it all in post-processing. The astro CCD I own also has quite abysmal ISO (white light including infrared it comes out at ISO400, with a colour filter only 80). When I tried doing the same with post-processing D30 RAW files I failed miserably, just like somebody has stated. Also I established for myself that for extremely low light shooting it is better to use a D30 at ISO1600 (less noise) than at ISO100 with a 16x longer exposure. So the on-chip extra amplification that Canon and others put in does something beneficial. What or how I am quite unclear, so I remain bothered. I still do not see how using a higher ISO setting (ie higher on-chip amplification) can avoid shrinkage in the dynamic range available at lower ISO. Ignore the special ISO 50 setting in a 1Ds, which seems to have less dynamic range than the ISO100 setting. I have seen some reports that extremely aggressive contrast stretching (those astronerds again!) throws off the colour balance in the darker parts of an image. While I have not really noticed this, it may be one reason why digicams still retain those ISO settings. But again, CCDs are very linear in response, so why should low light (low electron well-fill) colour balance shift? So I remain bothered regarding my deficiency in understanding what is going on...

[BJL]

The latest information from Rainer, samirkharusi etc. makes me think that the high ISO camera settings do more or different digital noise reduction processing than lower ISO settings. I do not see how the reported noise advantage of ISO400 could be just reduced discretization errors due to the A/D converter receiving higher level inputs due to more analog gain, because my reading suggests that fluctuations in electron counts due to thermal noise and such dominate over discretization errors at all signal (brightness) levels.

Could in-camera noise reduction do better than postprocessing with software like Neat Image, due to the camera "knowing more" its own noise behavior, like monitoring its own "dark noise" characteristics?

It seems likely that I had overlooked some possibilities in my original comment, and that the purest "do it yourself" approach of minimally processed RAW output plus more off-camera postprocessing is only worthwhile in special cases where data integrity and maximum operator control justify significant extra work: probably astrophotography, so I think I understand the lack of ISO adjustments there; and still maybe some high contrast scenes where highlight blowout is a risk. Most of the time, a histogram check or experience with one's camera's highlight handling should make it safe to rely on in-camera processing.

Oh well, at least I might have given Rainer some new aspects to study as he develops his skills with his new photographic tool! (I look forward to the resulting desert photos on the web sites.)


P. S. At least samirkharusi seems have to corroborated me on the low inherent sensitivity (ISO equivalent) of electronic light sensors! Perhaps the ISO should sue over the misuse of their name.

[Ray]

Could in-camera noise reduction do better than postprocessing with software like Neat Image, due to the camera "knowing more" its own noise behavior, like monitoring its own "dark noise" characteristics?

BJL,
Absolutely! I think one can safely say this must be true. It's like the difference between a specialist and a generalist.

What happened to your idea of getting a 10D - the lowest noise camera in the history of DSLRs?