For which scenes does high ISO not matter so much?

[Redbriars]

I went out into the garden this morning and set my Nikon D50 on a tripod and took shots of the same lily at many possible combination of settings for ISO, file size, aperture. The lily was just being lit by the sun over the trees but the background was still in shadow. (Nova Scotia, Canada)

To my surprise even the highly compressed jpegs looked indistinguishable from least compressed ones, and the high ISO images looked fine also. Mind you, these were viewed on a screen and not printed.

This image was taken at ISO 1250 (set at 1200, which is the highest for a D50) and cropped to about 1700 pixels wide and then resized to 700 pixels for the upload.

Two questions:

Are there images that would have been much more degraded ? Darker images, or a busier scene, or perhaps a uniform background? And would the degradation be more obvious if the image was printed?

And second, does anyone know how Nikon creates the smaller images? The large image is 3000 x 2000 pixels but the smaller ones are half that or less. Does the camera do a resize internally before writing to the card? In other words, is the camera discarding some information to create the smaller image or is it just collecting less information from the start?

Many thanks . . . Alan

My image did not appear. I will try to find it.

[Mike Buckley]

It is often written that the higher the ISO, the less the dynamic range. I haven't noticed it but I also haven't empirically tested it and probably never will. That's because it makes so much sense to shoot at the lowest ISO value that is practical given a particular circumstance and to then deal with whatever the characteristics of the capture happen to be.

Noise is most easily created in areas of the image that are out of focus, dark, and underexposed though any area of an image that has any one of those characteristics can exhibit noise. The higher the ISO value, the greater the chance that noise will be displayed in those areas.

The important corollary is that a sharply focused, "properly" exposed image captured at a relatively low ISO value will tend to display little or no noise.

As far as printing is concerned, the likelihood of noise being displayed increases as the size of the print increases and as the distance between the viewer and the print decreases. As an example, I don't bother doing any noise reduction so long as I can't see noise when the image is displayed at about 33% or smaller. That's because most of our image viewing is at smaller sizes; if I display at a larger size that displays unwanted noise, I'll deal with it then but only then.

I don't know about the internal process used by a Nikon camera to generate a file size that is smaller than the largest possible size. Good question about that!

[Inkanyezi]

There is much confusion about this issue, and it is aggravated by software having an "exposure" slider, which invites further confusion.

Any chip has noise, and we often speak about a "noise threshold", which is the region of exposure where the subject matter and noise reach about the same level. Under this "noise threshold" detail in the image is lost, and above it, there is more subject matter than noise in the image. However, there is always more noise in the more exposed parts, but more exposure tends to hide it below what we regard as visible. Mostly we don't see that noise, unless the tonality is very even, which tends to make it a bit more visible.

Generally, noise in an image is not a problem. If you worry about noise, the real problem is another.

The noise thus is more prominent where there is little exposure, which will be the case when you "raise ISO", by amplifying the signal along with any noise. This will be benign as long as your subject matter falls within the dynamic range of the image that is produced by the chip. There will be slightly more noise toward the middle tones, but generally there will be no objectionable noise as long as the image is registered with pleasing contrast without highlights burned out. The latter problem is more worrying than noise. If you screw up your amplification too much, you might push the highlights above what can be registered by the chip. If you then try to save them by exposing less, important image tones might become drenched in noise, because your dynamic range is insufficient. As a rule of thumb, you lose one f-stop of dynamic range for each full stop you raise the ISO number.

So whether noise is a problem or not depends a lot on subject matter and contrast. If you have much tonality you wish to retain, as well as detail, in the shadow, it is unwise to try very high ISO settings. If however, your subject matter is rather flat contrast-wise, and there are no prominent highlights where detail must be rendered, you can raise ISO substantially, without getting objectionable noise in the image.

Regarding the question of small jpeg images, they are sampled and shot with full resolution, but downsized in the jpeg conversion, so all information is collected, but whenever a jpeg is made, much information is jettisoned. A smaller jpeg will display less noise, exactly because of this process. I don't know your camera model, but generally, you cannot shoot a smaller RAW image with less resolution. In those camera models where you can shoot a smaller RAW image, most, if not all, will crop the image. There should be information about this in the manual for the camera.

[Shadowman]

Try including the sky in a shot with those same settings or any image with some deep shadows.

[Manfred M]

Allan - for technical reasons, darker parts of the image, especially in the blue channel are the ones more prone to noise. From a practical standpoint, we are more likely to notice the noise in large dark areas, even though there can be a fair bit of noise in smaller areas. We just don't notice this as much. Night skies and deep shadows are often where we see a lot of noise.

Camera manufacturers have done a lot of work to improve low noise performance, so newer cameras with properly exposed images will likely show a lot less noise than your D50 will, even at much higher ISO settings.

The fact that you can't see any difference between the different quality settings on your camera is not necessarily surprising. First of all, most computer screens display roughly a 2MP image, so anything that you do see will be down sampled from the D50's 6MP to fit on the screen. Depending on a number of factors (including the size and quality of your screen) you might not see any difference at all; BUT if you start editing the images, issues like compression artifacts and blocking are more likely to show up in the highly compressed files much more easily than with the ones with less compression.

Smaller images are created by down sampling (nicely said, throwing data away).

[NorthernFocus]

Others have already explained the noise phenomenon and downsampling. Even at the small size of the image as posted there is in-fact a good bit of noise evident in the BG shadows which demonstrates Manfred's point that noise manifests itself worse (visually) in darker/underexposed parts of the image.

[xpatUSA]

For which scenes does high ISO not matter so much?

Alan, welcome to the forum . .

In addition to what has already been said, one general answer to your general (title) question is:

High ISO settings do not matter so much for scenes which have adequate lighting and suitable content for ISO 200 (D50 base) but which, for artistic reasons, necessitate a very high shutter speed to freeze motion or a very big f-number for extreme depth-of-field.

[dje]

Hi Alan

Increasing ISO effectively means you are reducing the exposure experienced by the sensor and this is then compensated for by amplifying the signal out of the sensor. Probably the most common reason for doing this is to get a faster shutter speed.

Higher ISO means more background noise, which can be seen in the darker areas of an image, particularly if you lighten up these areas in pp. Higher ISO also means lower signal to noise ratio in all parts of the image, including the brighter areas. You can see this in areas where there is very little tonal variation eg a clear blue sky.

Noise reduction is usually applied in software for higher ISO images - either in camera for the production of in-camera jpegs or in raw processing software if you are shooting raw. In both cases, there is a trade-off between noise visibility and image sharpness. So there is no free lunch and the moral of the story is to shoot at the lowest possible ISO consistent with the shooting situation.

Dave

[Timar]

Further complicating the matter is that different sensors show very different noise characteristics.

Everyone who has ever tried to correct an underexposed shot taken with a P&S or mobile phone camera knows that those small sensors produce much more noise than the larger sensors found in DSLR or mirrorless cameras. This is simply due to the physical limitation of their small size - the smaller the light-gathering photosites, the less photons they are able to capture at a given ISO value and the more the remaining signal has to be amplified, which also amplifies the noise floor.

Now the noise you are seing is either this "physical noise" that is inherent to the quantum properties of light and matter as such and thus cannot be avoided, or it is "electronic noise" that is not due to physical limitations but originates from imperfect electronics within the camera, such as the noise introduced by the A/D converter. Usually, what you are seing is a combination of both.

However, the lower the ISO value and hence the "physical noise", the larger the proportion of "electronic noise" present in an image, although modern sensors and camera electronics have come a long way in reducing the latter kind of noise. In fact recent generation Sony sensors (employed by Nikon and Pentax) have gone so far as to almost entirely eliminate "electronic noise".

In practise this means that those cameras become essentially ISO-less at their base ISO of 200. Hence, for such cameras, manually setting an ISO value essentially becomes a relict from the times of analogue film. The camera could as well always expose at ISO 200 and digitally correct the exposure internally - just as you do when correcting the exposure of a RAW file in Lightroom.

But what does raising the ISO mean for a digital camera anyway? It simply means that the signal from the sensor - which always works at a base ISO just like a film - becomes amplified before the A/D conversion step. This is necessary only if the A/D converter is the bottleneck limiting the dynamic range that can be captured. For ISO-less cameras, the converter has a sufficiently low noise floor to avoid any bottleneck effect and the same amplification can be done digitally after conversion.

All these technical considerations aren't significant as long as you shot JPEG and don't try to capture HDR scenes by adjusting the exposure in Lightroom. Just let the camera automatically choose the ISO and don't waste a second thought about it. If you shot RAW, however, it helps to know at which ISO value your camera becomes ISO-less - because if you select a higher ISO, you are essentially wasting dynamic range in the highlights without further lowering the noise floor in the dark parts of the scene.

My Canon becomes ISO-less at ISO 1600, so I never choose a value above that. In the rare circumstances when there is no way to get more light on the sensor, I rather underexpose by one or two stops (e.g. corresponding to ISO 3200 or 6400) to capture the optimum dynamic range.

[dje]

Hi Timar and welcome to CiC.

I agree with all that you say except for the last sentence of your post. I would suggest that if you are under exposing by one or two stops, you are not getting the benefit of the higher DR at ISO 1600 anyway.

Dave

[xpatUSA]

I would suggest that if you are under exposing by one or two stops, you are not getting the benefit of the higher DR at ISO 1600 anyway.

Dave

On the other hand, if the Canon is truly ISO-less at 1600 and above, the raw data for ISO 3200 at 0 EV is no different to that for ISO 1600 at -1 EV; equally ISO 6400 at 0 EV and ISO 1600 at -2 EV. And, if the captured raw data is the same, would not the captured dynamic range be the same . . . . or does the converter apply ever-increasing noise reduction for settings higher than 1600? Just asking, I know very little about Canon stuff.

[dje]

On the other hand, if the Canon is truly ISO-less at 1600 and above, the raw data for ISO 3200 at 0 EV is no different to that for ISO 1600 at -1 EV; equally ISO 6400 at 0 EV and ISO 1600 at -2 EV. And, if the captured raw data is the same, would not the captured dynamic range be the same . . . .

Yes Ted that's what I am trying to say, but you put it more clearly.

Dave

[Timar]

Hi and thank you!

I agree with all that you say except for the last sentence of your post. I would suggest that if you are under exposing by one or two stops, you are not getting the benefit of the higher DR at ISO 1600 anyway.

In the ISO-less range, the dynamic range decreases lineary with higher ISO values. For recent Canon sensors (5D MKII, 7D, 550-760D, 70D) this linear range begins at about ISO 1600. Therefore, increasing the ISO value from 1600 to 3200 decreases the captured dynamic range by one full f-stop. Instead of shooting at 3200, one can as well underexpose at 1600 by one stop (push development) and obtain the same dynamic range plus one f-stop of additional headroom for the highlights.

See this excelent article for an in-depth explanation illustrated with a nice graph of the 5D MKII sensor performance.

[xpatUSA]

Hi and thank you!



In the ISO-less range, the dynamic range decreases lineary with higher ISO values. For recent Canon sensors (5D MKII, 7D, 550-760D, 70D) this linear range begins at about ISO 1600. Therefore, increasing the ISO value from 1600 to 3200 decreases the captured dynamic range by one full f-stop. Instead of shooting at 3200, one can as well underexpose at 1600 by one stop (push development) and obtain the same dynamic range plus one f-stop of additional headroom for the highlights.

See this excellent article for an in-depth explanation illustrated with a nice graph of the 5D MKII sensor performance.

A good link that, I like Clark's stuff, he does at least put some work into his articles!

And, with so many definitions of "dynamic range" floating around these days, I think I'll read it again

All my Sigma cameras, past and present, are truly ISO-less. No "exposure triangle" for me, they never change from ISO 100 - the ISO "knob" being rusted into that position. Makes life very simple for this old man.

[dje]

Hi and thank you!



In the ISO-less range, the dynamic range decreases lineary with higher ISO values. For recent Canon sensors (5D MKII, 7D, 550-760D, 70D) this linear range begins at about ISO 1600. Therefore, increasing the ISO value from 1600 to 3200 decreases the captured dynamic range by one full f-stop. Instead of shooting at 3200, one can as well underexpose at 1600 by one stop (push development) and obtain the same dynamic range plus one f-stop of additional headroom for the highlights.

See this excelent article for an in-depth explanation illustrated with a nice graph of the 5D MKII sensor performance.

Hi Timar

I'm not disputing the relationship between Dyamic Range and ISO setting. You can see similar curves to those in your reference in the data on the DXOMark site. As you are probably aware, the current Nikon cameras show DR decreasing in pretty much a linear fashion from base ISO up (sorry, an unkind dig from a Nikon user !).

What I am questioning though is your statement "obtain the same dynamic range plus one f-stop of additional headroom for the highlights". This I don't follow. For me DR is the ratio of maximum signal able to be recorded to the minimum signal able to be used (which basically means the read noise or there-abouts). The highlights are the highest signal area and represent the max signal that can be recorded. If you under expose by 1 stop at ISO 1600, you will get the same dynamic range in your capture as you would for full exposure at ISO 3200. This is one stop less than the max dynamic range available at ISO 1600 but if you under expose by 1 stop you can't achieve that full DR.

Perhaps what you are saying is that with ISO 1600, you have 1 stop margin of error for the highlights if you get your exposure setting wrong. I think I can see where you are coming from now.

Dave

[Timar]

What I am questioning though is your statement "obtain the same dynamic range plus one f-stop of additional headroom for the highlights". This I don't follow. For me DR is the ratio of maximum signal able to be recorded to the minimum signal able to be used (which basically means the read noise or there-abouts). The highlights are the highest signal area and represent the max signal that can be recorded. If you under expose by 1 stop at ISO 1600, you will get the same dynamic range in your capture as you would for full exposure at ISO 3200. This is one stop less than the max dynamic range available at ISO 1600 but if you under expose by 1 stop you can't achieve that full DR.

I know it is sometimes hard to wrap your head around these concepts. Let me explain in more concrete terms: If you are entering the linear ISO-less range at ISO 1600 (see the Clarkvision article), then increasing the ISO from 1600 to 3200 reduces the dynamic range by one full f-stop. With ISO 1600, you may be able to capture 10 f-stops, with ISO 3200 only 9 f-stops remain. By underexposing and push-developing by one f-stop at ISO 1600, you are effectively using the lower 9 f-stops, corresponding to the DR available at ISO 3200. The remaining upper f-stop is either left unused (if your scene has a real dynamic range of 9 f-stops or less) or available for highlights (a common scenario for night photography: the dynamic range of indirect lighting is limited, but the direct light sources blow out).

[dje]

I know it is sometimes hard to wrap your head around these concepts. Let me explain in more concrete terms: If you are entering the linear ISO-less range at ISO 1600 (see the Clarkvision article), then increasing the ISO from 1600 to 3200 reduces the dynamic range by one full f-stop. With ISO 1600, you may be able to capture 10 f-stops, with ISO 3200 only 9 f-stops remain. By underexposing and push-developing by one f-stop at ISO 1600, you are effectively using the lower 9 f-stops, corresponding to the DR available at ISO 3200. The remaining upper f-stop is either left unused (if your scene has a real dynamic range of 9 f-stops or less) or available for highlights (a common scenario for night photography: the dynamic range of indirect lighting is limited, but the direct light sources blow out).

Timar I understand where you are coming from. The reason I referred to the exposure margin of error in my previous post is that the advantage you mention is only achieved if the exposure value used, assumed to be the same in each case, causes highlight clipping on the ISO 3200 image.

Dave

[Timar]

Timar I understand where you are coming from. The reason I referred to the exposure margin of error in my previous post is that the advantage you mention is only achieved if the exposure value used, assumed to be the same in each case, causes highlight clipping on the ISO 3200 image.

Ok, I guess we got our wires crosses because I am coming from the right (exposure-wise) while you are coming from the left. Of course, if you plan to brighten up dark areas in Lightroom and you don't expect highlights to be a problem, you would be better off exposing normally and thereby increasing the dynamic range in the shadows.

Adding to the confusion is the fact that those terms are often used ambiguously. Dynamic range, for example, can mean the absolute S/N ratio, or it can mean "usable" dynamic range, which is limited by the perceptual quality of the noise floor. On the other side, a certain amount of clipping in the highlight may not be objectionable (even Lightroom allows some small-scale clipping before showing the highlights warning), particularly for direct light sources. Hence when discussing dynamic range, one can easily be at cross purposes.

[dje]

Yes some people like to make the distinction between "engineering DR" (referring to the ratio of max signal to noise floor) and "photographic DR" for which the lower end is defined by a certain signal to noise ratio around the 2 or 3 db mark.

[xpatUSA]

Slightly OT but over on the Sigma Forum controversy rages perennially about what is the "best" ISO setting for the Merrill series of cameras. The consensus is ISO 200 and that is mostly based on more headroom being available and they "ooh and ahh" about highlight "recovery" etc., etc. I've done the calc based on the ISO saturation method and it's more like ISO 160 since that method does include a defined headroom.

Having said all that, the slightest application of EC in any form makes all that stuff moot because the final arbiter of exposure is the scene itself and how one wants to render it, luminance-wise.

Here's a couple links by some well-respected folks:

http://theory.uchicago.edu/~ejm/pix/.../noise-p2.html

http://dougkerr.net/Pumpkin/articles...amic_range.pdf

Not forgetting CiC of course:

https://www.cambridgeincolour.com/tu...amic-range.htm
.