5D native ISO: what is it, does it matter?

This intriguing article:

http://www.digitalphotopro.com/articles/2006/mayjune/isospeeds.php

explained that digital sensors have native ISOs, and that when one tweaks the ISO setting, "amplifiers in the image sensor’s circuitry increase the gain before sending the image data to the A/D converter to be digitized."

Does anyone know the 5D's native ISO? And does it matter at all, in any way?

(I've been curious since I saw this article, and thought that this excellent new forum would be a good place to ask.)

Just a wild guess here, but wouldn't that be the lowest ISO speed the camera is capable of? Unless they use attenuators as well of course.

I think "native" ISO are 100, 200, 400, 800, 1600.
I know the new Canon dslr's allow ISO to be increase in third of each stop/setting.
The ISO can also be expanded to cover ISO 50 and 3200 as well. The 50, I believe, reduces dynamic range of the shot but can be helpful in really bright situations where you are getting the right side clipped. The ISO 3200 is actually a digital trick and not a real ISO setting if memory serves me right.

You get the native speed (and the best image quality) of the 5d at 100ISO and the real speed at this setting is about ISO125.

The later is only important if you use handheld meters otherwise you don't need to bother about it.

This all depends on what you mean by "native ISO sensitivity".

If the implication is that the sensor itself has a native sensitivity, the result of some unique gain of the amplifiers that precede digitization, leading to a unique sensitivity, and that we then change the gain of the amplifiers to depart from that, this is a misconception. There is no magic gain value that we can depart from. There is no sensitivity of the photodetectors until they have been provided with some gain (which could be 1.0) between their output and the input to the analog to digital converter.

In that vein, if a sensor has amplifiers that can be set digitally to give (only) gains that result in the ISO sensitivity being, say, ISO 100, ISO 200, ISO 400, ISO 800, and ISO 1600, then we could (if we felt a real need to use the phrase) say that all of these are the "native ISO sensitivities" of the sensor.

Now, in a particular camera model, the manufacturer may elect to provide ISO sensitivities other than those provided by the available repertoire of amplifier gains by scaling the digital outputs, using digital arithmetic, by various factors, thus establishing intermediate ISO sensitivities. It appears as if the intermediate sensitivities of the EOS 30D are attained that way.

Those intermediate sensitivities have some disadvantages with respect to saturation level and the like (perhaps leading to compromise in dynamic range or "posterization"), which have been discussed at some length in various forums.

Perhaps one could choose to say that these ISO sensitivities are not "native sensitivities" of the sensor. I find no attraction in the phrase.

A distantly related issue is whether the various ISO sensitivities are stated in accordance with the ISO standard for determining and stating the ISO sensitivities of digital cameras. As has been mentioned here, it appears that the sensitivity designated by Canon "ISO 100" would actually be properly described as perhaps ISO 125 or ISO 135 if determined and stated in accordance with that ISO standard.

I presume that this is a means of Canon having their cameras depart form the "standard metered exposure" prescribed by the interaction of two ISO standards, that for the "calibration" of an in-camera exposure metering system and the one for the designation of ISO sensitivity.

The doctrine set forth by the interaction of these two standards allows about 1/2 stop "headroom" to avert overexposure of the highlights in the case of a scene having a low average reflectance (assuming that the metering system is a basic ("dumb") "average luminance" system.

Very likely, since the centerpiece of Canon's metered performance is their Evaluative metering scheme, which uses an intelligent algorithm to predict, among other things, what the greatest luminance in the scene probably is, Canon may feel that this "headroom" is wasted. They could crank that out by changing the "calibration" of the exposure metering system itself. But if they did that, then photographers using an external exposure meter (one calibrated to ISO standards) would get a different exposure recommendation than that which would be given by the camera's metering system, a sure cause of consternation.

But instead, Canon could have chosen to understate the ISO sensitivity of its sensor, The result would be a standard exposure doctrine that "recovers" the headroom, while keeping apparent consistency between the cameras; metering system and external meters.

That is of course just conjecture on my part.

Having always had problems calibrating my canon cameras to hand meters I went to a local large store and tried out 8 meters in incident mode, they all agreed to within 2/10 of a stop. I then shot using that metered setting with my 5D and their rental 1D mkII, 1Ds and 1Ds mkII and every single one was consistently underexposing by 1/2 a stop. The histogram showed that the incident metered whites in the frame didn't even get into the fifth (rightmost) box of the histogram. I now set my meter to iso 80 when the camera is at iso 100 and get consistently very slightly underexposed images, to preserve the highlights. The cameras meters of course were accurate with the 1 series having the edge with accuracy on the 5D.

I know how to meter (incident metering does not need to adjust for blacks/whites!) and those cameras were all underexposing by half a stop. ISO 100 on canon cameras is more like iso 80-64 from all my testing and from the 30,000+ frames on my 5D, 25,000 on my 1Ds and a good 40,000 shot on the 10D. Makes me wonder about all those reports of the 5D's iso 100 really being iso 125, you must be joking!

Ben,
do you remember the heated debates about the 'real' speed of Velvia film? With digital the spectrum is even larger as a lot of different workflows have substituted the standard E6 process.

For example for RAW conversion I prefer Canons DPP. This SW is perfect for handling shadows but can be tricky with highlights. ACR has a contrary profile delivering better highlights but worse shadows. I can offer no formal tests but would not be surprised if the choosen RAW converter alone takes care for a difference of 1/2 to 2/3 f/stop.

Olaf

Hi, Ben,

Makes me wonder about all those reports of the 5D's iso 100 really being iso 125, you must be joking!

Click to expand...

My conclusion in that regard is solely based on my analysis of a test recommended by Chuck Westfall of Canon USA for confirming that the exposure metering calibration of an EOS-series camera was as intended by Canon.

Best regards,

Doug

Olaf_Laubli said:

Ben,
I can offer no formal tests but would not be surprised if the choosen RAW converter alone takes care for a difference of 1/2 to 2/3 f/stop.

Olaf

Click to expand...

Hi Olaf,

I did some semi-formal tests that show that C1 renders 1/3 stop to 1/2 stop brighter than Canon or ACR: conversion histograms

In fact, way back in v1.1 in response to vocal complaints Phaseone altered C1 to more closely match Canon's. Apparantly I was even more vocal because they shortly restored the old rendering:

>Don,

>Yes. Old curves are going back in.
>Some people will be annoyed by this move...

>This falls into the "Dont fix it if it aint broken" catagory
>... happens waay too often in software development.

>Regards,
>- MichaelJ

DL
>>The old curves simply are a better match for what I see when spot
>>metering and what I see on the camera review histo.

MJ
>>>Actually we did intentionally lower the exposure for 1D/1Ds
>>>back in February image because people was complaining about C1 being
>>>1/3 f-stops too high compared to Canon FVU.

- DL

Hey guys I'm just saying what I've been seeing, the 1/2 a stop is based on both the in camera histogram and ACR. In DPP I have to admit it's even worse.

The trick is to work out what the real value is on your camera and hand meter accordingly. I find it somewhat annoying though that the great noise at higher iso's is compromised according to my findings, it isn't such a high iso in reality after all!

I'm sure others will have different opinions but when an incident metered white doesn't even touch the rightmost part of the histogram and the meter is showing the same reading as 8 other meters from 4 different makes...

> The trick is to work out what the real value is on your camera and hand meter accordingly.

Exactly Ben, back in film days I would routinely alter the ISO setting on my meter to get the results I wanted. Unfortunately there's no adjustment on the internal meter (hence my issue above) - with an external meter I would simply tweak the ISO.

> I then shot using that metered setting with my 5D and their rental 1D mkII, 1Ds and 1Ds mkII and every single one was consistently underexposing by 1/2 a stop.

Note that C1 would probably render as you expected.

- DL

What meter do you guys recommend from todays offerings ?

The last hand held meter I used was a Gossen LunaPro in the 70's with my film gear in the 70's. Did not have a meter in the camera then. The Gossen was very accurate (at least my fading memory of it).

harvey

Hi Doug,

Your informative post got me thinking (which is often dangerous) about ISO and sensor gain. Would it be possible for the camera's CPU to use several exposures to calculate an appropriate gain for each photosite? This would effectively set to per-pixel ISO, which is a strange concept.

For the sake of argument, let's assume a bit depth of 8, so the possible values for any photosite range from 0-256. The camera would take one exposure, and certain photosites would be clipped at one end or the other. The camera would then adjust the gain at each of the clipped photosites and re-expose. In a four-pixel image exposed three times, the matrices would look something like this:

(numbers in parentheses indicate per-photosite gain expressed as ISO)

Exposure 1:
--------------------
233 (400) | 256 (400)
000 (400) | 125 (400)
--------------------
Exposure 2:
--------------------
233 (400) | 250 (800)
000 (200) | 125 (400)
--------------------
Exposure 3:
--------------------
233 (400) | 250 (800)
020 (100) | 125 (400)
--------------------

This would effectively create an HDR image straight from the camera, which is not always a desirable thing, but it could be useful under some circumstances. Of course, it may be computationally simpler to simply composite the three exposures in a manner similar to that in which HDR images are created in Photoshop.

This idea assumes that:
(a) the gain-setting process happens before the shutter is opened and
(b) it is possible to apply gain to individual photosites.

I believe (a) is true, though I could be wrong, and I have no idea whether (b) is true.

I suspect this is either an old idea or something that won't work in principle. Do you have any thoughts?

Thanks very much,

Jason

FWIW, this paragraph from the linked article prompted my thread:

Image sensors have an innate “native” sensitivity, generally in the ISO 100 to 200 range. When you set a higher ISO speed, amplifiers in the image sensor’s circuitry increase the gain before sending the image data to the A/D converter to be digitized. The sensor’s sensitivity doesn’t actually increase; the camera is just amplifying the data it produces.

Thanks to those who have responded.

> This would effectively create an HDR image straight from the camera

There's already prototype non-linear sensors and Canon has patents on other methods for HDR also.

- DL

anthros said:

I suspect this is either an old idea or something that won't work in principle.

Click to expand...

In fact something like that, although statistically a bit more robust by averaging (10 or more shots is common), is used in astro photography. It is common to calibrate for systematic sensor induced noise, both at the dark end (dark field subtraction) and at the light end (flat field correction, also adjusts e.g. vignetting).

Bart

Don and Bart,

Thanks for your replies. The nonlinear sensor article was really interesting. I'm not surprised to learn that astrophotography does something similar to what I've proposed. I once took a tour of the Keck observatory on the Big Island of Hawaii, which impressed me to no end.

The more I think about it, the less sense it makes to use this sort of thing to increase dynamic range, as post-processing with several images can do the same thing perhaps more easily. However, it would be nice to know whether its feasible to adjust ISO (i.e., sensor gain) on a per-photosite basis.

That would let you do other interesting things, such as varying the ISO linearly (or according to any function, for that matter) from the bottom to the top of the frame. This would effectively simulate a graduated neutral-density filter. You could also vary ISO radially from the center of the frame, correcting for a known vignetting problem (though this is perhaps better done in post-processing; however, it also sounds similar to what Bart mentioned).

For what it's worth, I'm a mechanical engineer. I say this not to lend credence to my ideas but rather to make it clear that I am not laboring under the misconception that I am an electrical engineer.

Cheers,

Jason

There's only one native ISO.

For the Canon CMOS sensors it is nominally 100. For the CCD 1D it was 200.

The native ISO of an image sensor is the equivalent ISO sensitivity of the analogue light to voltage converter. That converter can't become less sensitive or more sensitive to light, just as film cannot, but we can and do control the amount of light it is exposed to. The converter produces a voltage that corresponds to the level of light received for each pixel, but that voltage includes valid signal data and unwanted electronic noise data. That voltage is then the source of all data that the sensor can produce for an image file.

Any scaling of the native ISO whether digitally or analog will scale the noise data as well as the signal data bacause they have already been combined in the original voltage outputof the light-to-voltage converter.

We can pretend we are using a higher ISO by underexposing the ISO 100 sensor as if it was working at the higher ISO, and then fiddling results. Taking the underexposed ISO 100 data and scaling it up to simulate higher ISO sensitivity increases the noise too. Scaling in itself need not change the signal to noise ratio, but as we have underexposed the original image by letting less light (signal) onto the sensor, we may have decreased the SNR prior to the data being scaled. Thus we may have a lower SNR as well as higher noise.

So yes, native ISO does matter. In summary:

Using the native ISO will always give the cleanest full-dynamic-range data files and images.

Higher ISOs will increase the absolute noise levels and may reduce SNR, but they still have the full dynamic range technically speaking.

Lower than native ISO will be cleaner than native ISO because the noise is reduced along with the signal, but the dynamic range is also reduced.

My conclusion in that regard is solely based on my analysis of a test recommended by Chuck Westfall of Canon USA for confirming that the exposure metering calibration of an EOS-series camera was as intended by Canon.

Click to expand...

For RAW or in camera JPEGs? For RAW, we's want to "expose for the right" once we know the actual ISO.

I haven't done this test for my 5D yet but this is what I did for my older Rebel.

Use a good Incident meter in even flat light to get a reading at ISO 100 set on meter.

Shoot something as white as possible (paper would do but I use a white balance square with very high reflectivity).

Set camera in manual and bracket over and under by about a stop suggested by meter in smallest increments possible. Oh, I'm shooting RAW.

Bring bracket into Adobe Camera RAW, turn Auto settings off. Now I examine which white shot is as close to clipping without actually clipping (for the Rebel, at ISO 100, the white square read 250/250/250 and the next lighter bracket was blown out).

My conclusion (which could be full of holes) was that with this camera, shooting RAW, the lowest ISO really was 100. Not sure how the D5 will compare. So using my Minolta, I just set it at ISO 100.

Hi,
About the metering and settings on the meter.

For ISO50 I use the 50 setting on a Seconic L358
For ISO100 arround 80.
This will give me a red channel that is VERY VERY close to blowing out.
Using 125,00 as mentioned on some sites I will blow out the whites, although they can be recovered with C1 or PS.

You have to check the histogram for each channel and not the combined.
Also don't judge it by using the auto settings on for example C1 or PS, most of the time they will add 1 stop extra but washing out the picture.

Using the setting 80 for ISO 100 and pointing towards my lightsource (as you should do) I get 100% perfect exposures, no matter what light setup I use.

Greetings,
Frank

Tim Adams said:

Does anyone know the 5D's native ISO? And does it matter at all, in any way?

Click to expand...

The way I understand it, Native ISO of a camera refers to the externally metered ISO setting that result in 3.5 stops of headroom from the average metered value to the saturation point of the green channel, in the RAW data. Of course, one could speak of the native ISO of the sensor sandwich (sensor + filters), or even the sensor by itself (higher ISO). No one has ever met my request of a saturated ISO 50 RAW (over-exposed grey scale) from a 5D and a 100 with the same manual exposure to compare it to), so I can't tell you if the camera uses full sensor wells at ISO 50, or just pulls ISO 100 1 stop, and wastes 27,000 collectable photons.

I use ISO50 almost all the time in te studio and can tell you that if you use ISO50 and measure with a flashmeter and after getting perfect exposure with ISO 50 you use ISO100 you will blow out the shot.

So there is a difference.

What is true without a doubt is that highlights in ISO50 are more sensitive for clipping than on ISO100 and up.

But ISO50 measured perfectly is ISO100 blown.

Frank Doorhof said:

What is true without a doubt is that highlights in ISO50 are more sensitive for clipping than on ISO100 and up.

Click to expand...

That is expected, as there isn't enough photon capacity in the wells. There is only about 53% more capacity in the wells than is used for full RAW digitization at ISO 100 on the 5D and 1DmkII, as opposed to th 100% more needed for equal DR.

What I was wondering about, exactly, was how much (if any) of that 53% is used at ISO 50.

If the extra capacity is not used for ISO 50, then there really is no point in using ISO 50; you should just expose ISO 100 to the right, because readout noise is going to be far more than half of what it is at ISO 100, at ISO 50, and shot noise depends completely on the photons collected, not the ISO setting. Of course, your converter of choice may or may not have issues with extreme RAW highlights, even if they are not clipped.

But ISO50 measured perfectly is ISO100 blown.

Click to expand...

That doesn't sound like a lot of headroom is left at 100, either.

With my 10D, I could meter externally to ISO 32 in an area of even lighting with the camera set at ISO 100 and not blow RAW highlights if they are matte (the brightest pixels in a color-checker white square were almost clipping).

John,

This is O.T. Still, perhaps you know the answer! It's a question of individually addressing each sensel as a separate device.

Do you know of any available camera which can keep the shadow areas collecting photons after the brighter areas have reached a sufficient count so that the number counted is accurate and those sites are shut off?

This has been part of the promised CMOS design and delivered for testing but I wonder where we are with a chip where each pixel is independantly addressed?

Asher

@John,
I think you misread my answer.

When I measure for ISO50 the exposure is dead on.
When I set the camera to ISO100 with the setting for ISO50 the highlights are blown.

When I measure for ISO100 and shoot, the exposure is dead on.
When I set the camera for ISO50 with the measurement for ISO100 the highlights are too low.

I have read the conceptus that ISO50 would be exposing ISO100 to the right but I don't know for sure, when I calibrate my meter I take into account some headroom.

I would love to hear one definitive answer however from Canon

For the moment I keep using ISO50 because the noise is INCREDIBLE low in the shots and the color pops more than on ISO100, even when exposing to the right more than normal.

Greetings,
Frank

Frank Doorhof said:

When I measure for ISO50 the exposure is dead on.
When I set the camera to ISO100 with the setting for ISO50 the highlights are blown.

Click to expand...

That indicates that your ISO 50 exposure measurement (whichever method you used), achieved a near saturation based exposure. That will give the best signal/noise results for that ISO setting, as it should IMHO.

When I measure for ISO100 and shoot, the exposure is dead on.
When I set the camera for ISO50 with the measurement for ISO100 the highlights are too low.

Click to expand...

Again, makes perfect sense, although a question could be; how low are the highlights? I'm going to explore the ISO 50 characteristics a bit more myself, since it does indeed reduce the noise level by 10-12% relative to ISO 100, for my 1DsMk2 anyway.

I have read the conceptus that ISO50 would be exposing ISO100 to the right but I don't know for sure, when I calibrate my meter I take into account some headroom.

I would love to hear one definitive answer however from Canon

Click to expand...

While we keep hoping for a word from Canon, I am currently working on finalizing an evaluation of ISO 100, and ISO 50 is next !!! I expect the same principles will apply to other Canon camera models, but that can be verified once the benchmark is established.

My goal is to finally establish how much of a penalty (if any) ISO 50 really creates for Dynamic Range, or if it is just a matter of exposure measurement. In a controlled studio lighting environment the scene DR can be adjusted, but in outdoor situations it becomes much harder so one should be able to make a solidly founded decision without having to first build-up subjective experience. The foundation for that decision is taking shape ...

Bart

Asher Kelman said:

Do you know of any available camera which can keep the shadow areas collecting photons after the brighter areas have reached a sufficient count so that the number counted is accurate and those sites are shut off?

Click to expand...

I don't know of any (in current DSLRs) in the market, besides the dual sensel approach from Fuji (which is something different). Evaluations by myself, and others, sofar indicate a close to perfect linear relationship between exposure and resulting signal for DSLRs, and point and shoots. That does make processing that image data relatively easy. Most image processing benfits from linearizing/calibrating the response curve, which becomes unnecessary if it is linear from the start.

Non-linear sensel response can cause lots of complexity for color accuracy, although it could be done. My guess is that the cost associated with controlling that complexity is what is preventing it from leaving the Labs.
I'd expect a variable gain/ISO approach per sensel sooner than a variable anti-blooming kind of approach. Both avenues have their issues.

Bart

I will be looking forward to that evaluation.

Outside I use mostly ISO100 because the little bit noise is not a big deal but the highlights are.

If you are comparing also try the sharpening, on ISO50 I can sharpen much more without artifacts but that also be due to the lower noise.

Asher Kelman said:

John,

This is O.T. Still, perhaps you know the answer! It's a question of individually addressing each sensel as a separate device.

Do you know of any available camera which can keep the shadow areas collecting photons after the brighter areas have reached a sufficient count so that the number counted is accurate and those sites are shut off?

Click to expand...

I've read of such things being worked on, but not of any products. Of course, such a design would require knowing exactly *when* a pixel reached saturation, otherwise it is effectively clipped. If the total exposure was 1.53 seconds and the pixel saturated at 1 second, then we could assume that the pixel would have reached 153% saturation during the full exposure time.

This assumption requires a static scene, of course, not one where some parts are brighter in the earlier or later part of the exposure, for the assumption to be true.

This has been part of the promised CMOS design and delivered for testing but I wonder where we are with a chip where each pixel is independantly addressed?

Click to expand...

CMOS allegedly doesn't bloom, as currently implemented by Canon. What sometimes seems to be blooming may be out-of-focus IR light of high enough intensity to register in the scene, despite the IR-cutting filter.

When we see the first independently-addressable pixels, I would not be surprised to see lots of noise despite the high dynamic range.

Thanks John.

RE: O.T. CMOS with independantly addressed pixels

I think we will see a marked drop in noise since we wil have pixels shut of when they are say 97% full and the time that that took will be recorded and then adjusted accordingly.

One would merely set how much noise one wanted to accept and the exposure would be lenthned accordingly.

A "P" or program would be optimised for action to get the best noise exposure speed combo same with AV.

These CMOS chips do not need external shutter BTW, as the pixels are the shutters.

I don't know why one would expect more noise?

Asher