Auto ISO in manual = Shutter and Aperture priority.

[Ysarex]

Bumping the ISO will always reduce quantization noise as compared with adding back exposure in post. Unless the analog amplification is introducing enormous amounts of noise, bumping the ISO at the sensor will give you better results than digitally amplifying the signal in post.

No argument. All of my empirical tests verify what you just said there. We get less noise by raising the ISO -- I made the very point in my first post right? You did read where I said that right?

Hang on -- I'm working on posting an example with more info.

Assuming that you're not clipping anything that matters to you, bump the ISO.

This point about clipping does add an interesting twist because the typical amplifier in a camera will take solidly recorded data and gladly push it right into oblivion for you whereas had you not raised the ISO you could have kept that data.

Joe

 

[photoguy99]

No I have not read the whole thread.

I'm just not sure why anyone would rather work with a file that had the same data only with more noise. Honestly I'm wondering if either I missed a word (repeatedly, I did re read, but sometimes once you've missed a word you just can't see it) or if you didn't type a word. Possibly I simply have your meaning reversed.

Unless your point is mainly about clipping?

 

[photoguy99]

Oh wait, I did read that. Actually I take exception the the very next sentence. It does, or at least can, produce real data, to amplify the analog signal. Reducing noise, as opposed to removing it, reveals additional signal. Normally, anyways.

In fact, a standard noise reduction techniques is to simply increase the signal. See Dolby.

I look forward to tomorrow morning when I will look in to see your further details!

 

[Ysarex]

Oh wait, I did read that. Actually I take exception the the very next sentence. It does, or at least can, produce real data, to amplify the analog signal. Reducing noise, as opposed to removing it, reveals additional signal. Normally, anyways.

In fact, a standard noise reduction techniques is to simply increase the signal. See Dolby.

I look forward to tomorrow morning when I will look in to see your further details!

Hang in there, I'm working on the further details right now. I'll be posting an example in a few minutes but I want to post in response to Overread's comment. Bottom line is that once I have the more noisy non amplified sensor capture I can do my own noise reduction. Including if I want 5 different levels of different types of noise reduction applied selectively to different parts of the photo. The amplifiers can't do that. And I did say extreme more than once.

In originally posting here I wanted to make the point that raising the ISO doesn't permit recording more data. It just alters how the recorded data is processed. Amplification revealing additional signal? Nothing usable in my experience or tests.

Joe

 

[Ysarex]

yes, absolutely.



absolutely not.

Joe

Well that saved me a ton of time testing that theorem.. auto ISO it is! Lol

Exactly - if the amount of work in editing is increased significantly and if the gain is nothing to negligible I don't see any virtue in changing how everyone uses ISO currently.

No problem there; again I added this to make the point: Raising the ISO in a digital camera doesn't help or permit that camera to capture one iota of additional low-light data. It does all kinds of beneficial things like allow us to chimp the LCD, but it doesn't get us any more information. If that's the case I don't have to hate it but I can just ignore it. And maybe there's a benefit to ignoring it? I did say extreme.

Plus added to that everything I've read has always stated very clearly that a higher ISO gives less noise than underexposing using a lower ISO.

Absolutely, and I acknowledged that in my original post. If we amplify the sensor signal before A/D conversion we're going to get less noise up front. But, and this is camera make/model dependent, we're going to get different noise and noise uniformly applied.

Now I did say extreme. Let me repeat one more time that I used the word extreme in my first post about this. I'll admit to being an extremely obsessed pixel peeper here who's favorite quote is by Michelangelo, "Trifles make perfection, and perfection is no trifle."



So I did the same thing here that Braineack did. That's a screen shot from PhotoNinja's file browser. The first photo is for reference: ISO 200 properly exposed. In the middle file the ISO was raised to 1600 and you can see the shutter speed change. In the last photo the ISO was set back to 200 and the sensor was simply underexposed by 3 stops. In PhotoNinja the only thing I've done is what Braineack did in LR -- raise the exposure comp. 3 stops. PhotoNinja then adjusted the browser preview.

Now let's have a close look deep into the shadows. You need to examine this image full-res:

edit: I was afraid of this wouldn't get posted here full-res and it didn't. Here it is at full resolution: basket_noise.


The left image is again just for reference. You can see in the ISO 1600 image the reduced noise compared to the next image where the sensor was just underexposed 3 stops. Raising the ISO certainly has an effect and it certainly does deliver a less noisy result as you noted and as I noted. Examine how the noise is different though. Although the underexposed ISO 200 image is at first noisier it's noise that is very easy to filter. The next image over is pretty simply noise filtered. As good or even better than the ISO 1600 image?

Now we take it a step further. In post we always have the option to make local adjustments to our photos -- things like burning and dodging and so on. You can sharpen a photo in camera (how crude) but in post you can sharpen the midtones more and the shadows less and even use different sharpening algorithms on different parts of the photo. Did I mention extreme already? Well the same applies here. I don't have the option to just noise filter a photo I have the option to noise filter it in different ways to different amounts in different places. Trifles make perfection.

Now here's another interesting twist that came up in one of Photoguy99's posts. This photo was an extreme high contrast lighting condition. Note the window light in the upper left. In the ISO 200 correct exposure the window in the upper left has just begun to clip in the green channel. This scene was selected because it takes full advantage of the sensor's recording capacity. Now there really is nothing you can do to get a sensor to record more data than it can record. But if you're in a tight spot and you need as much data as possible be very careful raising the ISO. In other words what would you rather have; less noise and clipped highlights you can't fix or noise you can filter to have less noise? Now I didn't really say that because I would never want to advocate sloppy exposure.

In fact this theory goes against the tried and tested "expose to the right" theory.

Not at all -- it actually works really well in conjunction with ETTR theory. ETTR theory is all about capturing all the data the sensor is capable of recording. This is just another aspect of the same thing: If you think raising ISO is helping you capture more data, think again. That was the most important point I was making. And if that really is the case then ignoring it is no foul.

The OP was delighted at how freeing it was to just hand off the ISO chore to the camera. Derrel and Kristof and Todd all gave great examples of how valuable it can be to work that way. I agree. When I teach a first semester photo class I start them out with the ISO set to auto so they can concentrate on understanding what shutter and f/stop do. But as we continue and learn to process raw files and understand what a raw sensor capture really is I eventually take the ISO away from them entirely. They can have it back later, but it helps at some point to understand what's really going on. They get this strange notion (Lord knows from where) that raising the ISO is allowing them to record more information in low light situations. They need to learn that's wrong.

Armed with that knowledge and out with my little pocket camera that I always carry I find I need a faster shutter speed in low light. Do I raise the ISO or just twirl the EC dial and force the faster shutter speed? The only thing raising the ISO does is get me that faster shutter speed. Twirling the EC dial gets me the same faster shutter speed faster. The sensor is underexposed by the same amount either way.

Joe

 

[Vince.1551]

You guys should work for camera manufacturers

 

[photoguy99]

What Ysarex is doing will work pretty well, to be honest. He may find the results more pleasing, all the time, and that's OK.

Here's the thing. Quantization Noise (which isn't really noise per se, it's lost signal which we model as noise for most analyses because that works quite well) is usually associated with low spectral frequencies. That is, it's most visible is large more or less "flat" regions of very similar tone. It shows up as banding or posterization. In these cases, we've oversampled, we have tons and tons of pixels to play with in other words. We can fix this trivially. Ysarex example fixes quite a bit of it trivially and does a fine job. Anything with details that a are more than a handful of pixels across is going to be perfectly OK.

The places where it's less obvious are in areas where there is almost no tonal separation, but there's a fair bit of fine detail. Shadows are an obvious case. We could shoot something that comes out black at base ISO, but renders a (crummy) picture at ISO 1600. Still, this follows throughout the tonal range. Consider a cloud, with a lot of detail that's all very very much the same grey. Quantization noise will show up here by simply rendering entire regions as the same tone, losing the fine structure in the photo. It will actually look un-noisy. Noise removal won't do anything here, but the cloud is botched.

These are pretty subtle effects, and probably not much to worry about.

You can see this exact effect, I think, in Ysarex' example, and judge for yourself. Examine the band around the top of the basket. On the rear of the basket, it presents a dark, shadowed, vertical face to to. In a couple of places in the ISO1600 photo there are a couple of slightly lighter horizontal lines. These are pretty obvious in the correct exposure, but still visible in the ISO1600 photo. This detail is reduced to only the brightest of the lines, in the ISO 200 picture, and then largely obliterated by noise removal.

For a more middle range example, there's some sort of metal to the left of the basket, with floral decorations embossed (?) on it. Some of the petals have a pattern of parallel vertical lines, very subtle. Quite obvious in the correct exposure, IF you're looking for areas with fine detail contained in a very very narrow tonal range. The ISO1600 has a lot of noise in this area, but does preserve some of the feeling of those vertical lines, some sense that there's pattern in there. The underexposed photo loses all of that, and noise removal turns it into smooth mush.

These are very very subtle effects. 4 bits of quantization noise versus 1 isn't lot, and when the final is going to be crushed back down to 8 bits or so for viewing anyways, a host of sins get concealed. It's not the kind of thing I worry about personally in the slightest, but I am definitely not a pixel peeper by nature.

 

[Braineack]

Now let's have a close look deep into the shadows. You need to examine this image full-res:

e

Based on this I'd rather 1600; I don't use detail removal software.

Simply changing the ISO is easier, in my test there was no difference in quality between the shots, but you've demonstrated it's better.

 

[Ysarex]

Based on this I'd rather 1600; I don't use detail removal software.

Simply changing the ISO is easier, in my test there was no difference in quality between the shots, but you've demonstrated it's better.

Your test shows no difference and you get no argument from me. That's the big point I wanted to make in the first place. Put the esoteric and extreme processing aside. Photoguy99 is right -- at best we're talking subtleties that are only meaningful in the extreme. Todd asked if it really mattered and I said "absolutely not."

By saying you see no difference you're confirming what I was primarily interested in getting across. Variable ISO is not a capability of digital camera sensors. Their light sensitivity and recording capacity is fixed in manufacture. Raising the ISO on a digital camera underexposes the sensor period. All you get from the ISO increase is the JPEG support. The JPEG support is great but as I said, my students all get this idea that raising the ISO on their cameras is actually allowing the camera to record more low light data when in fact it's doing the opposite. It allows them to get a hand-holdable shutter speed or smaller f/stop but it's underexposing the sensor to permit that.

So I'm completely happy that your test shows that raising the ISO and underexposing the sensor produce no visible difference. If there is a difference it takes us into the esoteric realm of trying to sort out extreme differences in noise processing. I'll defer to photoguy99 as the electronics expert there. I'm the raw file processing nut job.

It's my job to make sure new photographers really understand what they're doing. To that end I spend a lot of time testing stuff. The extreme processing lunacy comes from my testing and also from the advantage that I get from getting my hands on all of my students different make/model cameras. During a semester if I want a T5i or a D5300 or even a D700 or a 5DmkIII I can usually just grab one from one of my students. So a few years ago when I first got my hands on NoiseNinja and tested it I was pretty impressed. Eventually I got around to testing what I could get from it versus the camera's analog amplification and often I'd end up deciding I actually preferred NoiseNinja without the ISO amp. Last year DXo released their Prime noise filtering algorithm -- very impressive tool. So I think that deferring the noise processing step to post and skipping the analog amplification can give me a slight edge sometimes (different camera dependent), but my answer to Todd asking is it something that should give him pause to consider remains: absolutely not.

Back then to my original post and the advantage of using auto ISO: The speed advantage you get by just letting the camera deal with applying the sensor underexposure for you is invaluable. I agree. I just wanted to point out that that is what we're doing, auto underexposing the sensor, and that if you want to just leave the ISO at base and not even bother with setting it to auto you'll really get the same thing -- at best no visible difference. So when you first responded to my post by asking, "So you just want everyone to shoot at ISO 100?" Well, if that's the base sensitivity of your camera sensor then as far as the usable data that the sensor is recording that is what you're doing. Signal amplification doesn't add data. Of course the data has to be processed; amplifying what you've got can be done pre-A/D conversion or post-A/D conversion but neither way gives you more.

I may be just an old man rambling on too much, but this all gets to one of my persistent concerns and a reason I jumped in to this thread. I get troubled when the language we use disguises what we're really doing. The "language of ISO" on digital cameras disguises the fact that any increase in ISO is equivalent to underexposing the sensor. I'm not saying we shouldn't do that, I'm saying we should understand what we're doing and call it for what it is. Correct understanding helps us make the best choices. I get a constant stream of students who are confused about what ISO on a digital camera really does and all we have to do is look around here on TPF to find more of the same. My students are shocked when I tell them that raising the ISO reduces the camera's capacity to record low-light data. The ISO language used in the industry has twisted their understanding a full 180 degrees.

Joe

 

[photoguy99]

I quibble slightly, I think it's more than simply JPEG support. I think the results are different, and that they preserve certain detail.

The detail that's preserved is pretty hard to retain. You can destroy it by accident any number of ways after the fact, and you're not likely to notice it.

Mainly, there's a slight difference in feel in the results, in at least some cases. Noise by itself can increase the feeling of overall sharpness, and the noise in play here is somewhat correlated with certain fine detail. So, if you're willing at accept some noise, you really are getting a sharper photo. Slightly sharper.

If you're allergic to noise, and that's not a judgement it's just a matter of taste, then it's going to matter less to you.

And, again, these are all very very subtle effects that don't matter if you're not printing large or otherwise digging quite deeply into the photos. I cheerfully bang the ISO all over the place, because I quite dislike mucking about with things in post and I trust the camera to be a decent job with the whatevers.

How detail oriented are you? What are your thoughts on ultimate sharpness, noise, and the tradeoffs between them? Choose your weapons wisely, now that you are.. exhaustively.. informed.

 

[SCraig]

To me the answer is simple:

ISO 1600 thanks to Auto-ISO jacking it up:

ISO 400 which is where I would have set it:

Identical processing on both, or as close as I could get with the crop. The shots were taken 31 seconds apart, the amount of time it took me to disable Auto-ISO and setup the shot again.

Just after daybreak and the bird was in a rather dim area, much like the majority of my shots. Auto-ISO jacked the ISO to 1600, the ceiling I had set, for 1/30 second whereas I would have shot at ISO 400 and settled for 1/8 second exposure. The first is a throwaway due to what I consider excessive noise. The second is not.

Oh, and I did NOT enable Auto-ISO again after I took the second shot. To each their own, I'm content with what works for me.

 

[photoguy99]

Rabbits are super slow, they are not fast at all! Why, I stuffed this rabbit into a sack and totally kicked its ass in a 50 meter foot race.

If you'd accept 1/8 why did you cripple auto ISO by restricting it to 1/30?

 

[SCraig]

If you'd accept 1/8 why did you cripple auto ISO by restricting it to 1/30?

I didn't. I set the ceiling to ISO 1600 so that's the shutter speed it chose at ISO 1600. I did say it was dimly lit.

After taking those shots I truly DID cripple Auto-ISO by turning it off entirely.

 

[Overread]

To be honest even at 1/30sec you're darn lucky to get a sharp shot of a living animal. Even a tiny motion will blur at that speed to say nothing of having a tripod or very steady hands (esp with a long lens).

Auto ISO I think did right in those conditions; you're just pushing the settings way beyond normal tolerances. Which is basically when most auto software fails.

 

[SCraig]

To be honest even at 1/30sec you're darn lucky to get a sharp shot of a living animal. Even a tiny motion will blur at that speed to say nothing of having a tripod or very steady hands (esp with a long lens).

Auto ISO I think did right in those conditions; you're just pushing the settings way beyond normal tolerances. Which is basically when most auto software fails.

That's my point. Much of the time I *AM* shooting in conditions where software fails so, despite all the accolades in this post, there are times when Auto-ISO is a detriment rather than an asset.

My personal opinion is that Auto-ISO is nothing more than a gimmick. I choose the ISO I want to use based on the light conditions at the time and don't want my camera arbitrarily jacking it up to some ceiling that I chose. If I'm going to do that then I'll set the ceiling at the ISO I want to be used and allow it to DROP the ISO but never raise it.

If I'm shooting in manual then I'm shooting in manual for a reason and I don't want my camera changing the settings. If I'm shooting in aperture priority or shutter priority then I have already chosen an ISO that I want used and allow the camera to choose the appropriate shutter speed or aperture. I have yet to run into a situation where I want the camera choosing more than one variable which is why I never, ever use full-automatic.

Again, my personal opinion and the way that I personally choose to do things.

Oh, and notice that the 1/8 second shot is sharper than the 1/30 second shot although both were shot at 500mm. I nearly always use a tripod. Once again, my personal choice.