What happened to low ISO?

[250swb]

I don't think there was ever going to be a 'visual revolution' with a few stops of improvement in high iso performance and no on promised one. You could shoot film at 6400 just like you can shoot a dslr at 6400 but what is does mean is that pictures that were once taken in these situations with loads of noise and fell apart at anything over 6x4" can now look very good printed to make larger sizes which is great for both photographers and clients.

Ask any good wedding photographer what high iso has done for them during church ceremonies where no flash is allowed.

Twas sarcasm that a visual revolution would evolve from buying an essential new camera......

For anybody that needs a new camera, try this crazy idea

Crazy Idea

Steve

[BernardLanguillier]

The truth is that we are still waiting to see some sort of visual revolution in images from high ISO performance that photographers said they needed so much before they bought their latest Nikon's and Canon's. That photographers now need low ISO performance is pretty well in the same ballpark, in that photographers soon forget all the things they needed and make the same old photographs anyway.

Your post reminded me of this excellent article.

Cheers,
Bernard

[KevinA]

This will give you ISO 6.

http://cgi.ebay.com/Kodak-DCS-Pro-SLR-n-Di...=item19baff5a63

You end up paying only 160 US$ per ISO, whatever that means...  

Cheers,
Bernard

I was about to mention the Kodak, I just sold one. The Kodaks low iso settings were not like conventional settings, you had a matrix of shutter speeds that had to be used. I don't know what Kodak did but the low iso settings produced wonderful results.
I can see an argument for 25 or 50 iso if they produced superior results, otherwise I can't think of a genuine need.

Kevin.

[fredjeang]

This will give you ISO 6.

http://cgi.ebay.com/Kodak-DCS-Pro-SLR-n-Di...=item19baff5a63

You end up paying only 160 US$ per ISO, whatever that means...  

Cheers,
Bernard

Hi Bernard,

What do you mean ISO 6 ??  

Is that Kodak really a good camera? it has been hardly critisized everywhere.

Fred.

[BJL]

Canon's ISO50 is fake ISO.

I understand that the ISO 50 has somewhat limited highlight headroom, so is "pulling" a bit further than is safe in some situations, though it seems fine when there is not too much subject brightness range between metered mid-tones and highlights: about 2 1/2 stops or less.

But can you explain the difference between "fake ISO" and "real ISO"? As far as I can tell, the only meaning of the ISO setting on a digital camera is exposure index, indicating the amount of light (subject brightness times exposure duration) needed to get a suitable "mid-tone signal level" in the final digital output. All ISO settings are equally real as exposure index values, as far as I can tell, even ones with limited highlight headroom or noisy shadows.

Some people seem to use "real ISO" for the effects of varying the gain in the analog domain, and "fake ISO" for similar effects achieved in the digital domain, as if variable analog gain (with its associated amp. noise) is "real" and "good", while variable digital scaling of levels (a perfectly noise free process) is "fake" and "bad". I do not see how this dichotomy is justified.

Each approach to adjusting the final signal level has its advantages, and my imaginary ideal camera would have a fixed amplification (enough to get the sensor dark noise level comfortably above the level all downstream noise sources), and an ADC that can handle that signal even with full-well output, and then ISO adjustment done digitally, to avoid any noise from variable gain in analog processing. So all ISO adjustments would be "fake"!

[Guillermo Luijk]

But can you explain the difference between "fake ISO" and "real ISO"? As far as I can tell, the only meaning of the ISO setting on a digital camera is exposure index, indicating the amount of light (subject brightness times exposure duration) needed to get a suitable "mid-tone signal level" in the final digital output. All ISO settings are equally real as exposure index values, as far as I can tell, even ones with limited highlight headroom or noisy shadows.

There is more to ISO than that. Look at the following diagram:



In the digital pipeline, the ISO adjustment controls the gain of an amplifier that works after the sensor captures light. Aperture and shutter control the amount of light (photons) that hit the sensor. ISO comes next, and controls the amplification of the analog signal obtained, and this amplification has very important consequences both in the signal to noise ratio and highlights headroom.

When setting ISO50 on a Canon 5D, the amplifier will internally work at the ISO100 gain. If you read carefully and understand the following, you will see clear that using ISO50 doesn't add anything to the RAW shooter: for a given aperture and shutter, the RAW data (i.e. the digital values contained in the RAW file) will be exactly the same, no matter if we set ISO50 or ISO100 on the camera.

It will be the camera display and commercial RAW developers that, according to the RAW file metadata informing about the ISO50 setting, will lie to us by displaying the ISO50 image one stop darker. But the RAW data itself will be exactly the same, both numerically and qualitatively (signal to noise ratio).

[Dick Roadnight]

Jeff,
I think we can acheive almost everything in digital.
My point was a question that reflects my ignorance on how to get close to the look of these Pan film with digital.
I know it can be done, I just can't find how.

Fred.

You can get the high noise-free res of pan films (I used Pan F in Hasselblads) with quality high-res MF digital cameras.

What limits ISO is the size of the pixel (or of the sensitive area or micro-lens of the pixel)...

So bigger pixels give you higher ISO.

Each pixel can sense a limited number of photons and record the signal from them before the signal store is full.

Smaller pixels give you higher res (per unit area) and lower ISO (as they sense fewer photons).

The smallest practicable pixel (pitch) is about 2 microns, as the wavelength of light is .4 to .7 microns.

So, if a 6 Micron pixel has a lower limit of ISO = 50, a 2 micron pixel, with 1/9th the area, might give you ISO 15 and the best res per area (and consequently the best DOF for the res).

In a very small pixel there is very little room for data storage, so the max ISO of a 2 micron pixel would make it unmarketable in this high-ISO mad world.

[fredjeang]

You can get the high noise-free res of pan films (I used Pan F in Hasselblads) with quality high-res MF digital cameras.

What limits ISO is the size of the pixel (or of the sensitive area or micro-lens of the pixel)...

So bigger pixels give you higher ISO.

Each pixel can sense a limited number of photons and record the signal from them before the signal store is full.

Smaller pixels give you higher res (per unit area) and lower ISO (as they sense fewer photons).

The smallest practicable pixel (pitch) is about 2 microns, as the wavelength of light is .4 to .7 microns.

So, if a 6 Micron pixel has a lower limit of ISO = 50, a 2 micron pixel, with 1/9th the area, might give you ISO 15 and the best res per area (and consequently the best DOF for the res).

In a very small pixel there is very little room for data storage, so the max ISO of a 2 micron pixel would make it unmarketable in this high-ISO mad world.

Dick,
If I understand correctly (I'm useless in maths and enginering) and resuming in practical: to acheive a 35mm film pan quality print we need now in digital nothing less than an MF ?

oh oh...if that is true, this would feed the debate about digi-film and 35mm vs MFD...  

Fred.

[Dick Roadnight]

Dick,
If I understand correctly (I'm useless in maths and enginering) and resuming in practical: to acheive a 35mm film pan quality print we need now in digital nothing less than an MF ?

oh oh...if that is true, this would feed the debate about digi-film and 35mm vs MFD...  

Fred.

You asked about "Pan film", and did not specify that you were talking about "35mm  Pan film".

When I gave up my film Nikon and moved to 120 film, initially I used 200/400asa film, but the results were not as good as Pan F 35mm, so I used pan F 120.

If 10 times the film size was the limit for enlarging from film, and if (for res better than the eye can see and less noise than the eye can see) you want no fewer than 360 pixels per print inch for enlarging from digital, then the best (60 Mpx) 645  MF digital is equivalent to 645 film.

35mm film would not be as equivalent to ff 35mm digital, because of the AA filters on DSLRs,,, but, if anyone ever made a quality ff 35mm camera, there is no reason why it should not be as good as 35mm film.

It depends which film you are comparing to which digital, and which yardstick you use (e.g resolution or noise).

If noise is your yardstick, Nikon is, I believe, a great deal better than Canon, and Hasselblad is a great deal better than Phase.

When you notice that a 2MPx 1080i TV screen looks as good as a 24 Mpx DSLR picture, it becomes apparent that it is not all about pixel count.

[fredjeang]

I was talking about low iso Pan films from 40 to down.
35mm gigabitfilm can resolve more than 400 PImm according to the french shops in Beaumarchais.

a link here:http://forum.getdpi.com/forum/showthread.php?t=7817

Fred.

[BJL]

There is more to ISO than that. ...

In the digital pipeline, the ISO adjustment controls the gain of an amplifier that works after the sensor captures light. Aperture and shutter control the amount of light (photons) that hit the sensor. ISO comes next, and controls the amplification of the analog signal obtained, and this amplification has very important consequences both in the signal to noise ratio and highlights headroom.

Your diagram fits my description, except that you then declare that only when the the signal level is adjusted in the analog domain is it "true ISO", while later adjustment of the digitized signal (multiplying the output numbers by an appropriate constant), is "fake". I am sorry, buy I do not accept that wording, given that the results either way are similar (though not identical) and I do not see that analog gain is unequivocally superior to later digital adjustment, so the use of adjectives "real" and "fake" seems prejudicial. Can you cite an official definition of "ISO speed" (from an ISO document, I suppose) to support this "fake" claim? To repeat, the only relevant definition I know of is an ISO definition of units of measurement of exposure index, which cover digital adjustments too.

I still see the strong possibility imagine that the best way to adjust exposure index while minimizing noise from the analog processing pipeline is to have fixed analog gain (as in a good power amplifier?) and an ADC of sufficiently large DR to cover from signal strengths from below the photosite noise floor to the full well signal strength, and then adjust the digital signal to the desired output level placement.

[BJL]

Smaller pixels give you higher res (per unit area) and lower ISO (as they sense fewer photons).

smaller maximum ISO speed before noise is unacceptable (to 100% pixel peeping at least!), but no change in minimum ISO speed, which seems to be the main topic of this thread.

The smallest practicable pixel (pitch) is about 2 microns, as the wavelength of light is .4 to .7 microns.

A dubious claim since the vast majority of sensors made these days have photosite spacing of less than 2 microns, and some are as small as 1.2 microns. It is not entirely clear than even the wavelength of light is an absolute minimum limit: antennas can be considerably shorter than the wavelength of the signal they receive. For example, AM wavelengths are about 200 to 400m, but most AM radios are smaller than that!

[Guillermo Luijk]

Your diagram fits my description, except that you then declare that only when the the signal level is adjusted in the analog domain is it "true ISO", while later adjustment of the digitized signal (multiplying the output numbers by an appropriate constant), is "fake". I am sorry, buy I do not accept that wording, given that the results either way are similar (though not identical) and I do not see that analog gain is unequivocally superior to later digital adjustment

If you think true and fake ISO yield similar results just look at the images on the left (shot at ISO100, numbers multiplied by an appropiate constant) vs the images on the right (shot at ISO1600), both at the same aperture/shutter (i.e. identical amount of light hitting the sensor in both cases):






The thick red curve in the following SNR vs RAW exposure plot shows SNR improvement when raising (true) ISO for a given aperture/shutter pair. In the deep shadows on a Canon sensor, SNR almost doubles (+6dB means doubling SNR) from ISO100 to ISO200, being the improvement less noticeable the higher the ISO setting. Going beyond ISO1600 doesn't produce any improvement, no matter if ISO3200 and above are true ISOs.




The 'true' or 'fake' ISO wording is just irrelevant semantics. The important thing to know is that there _is_ a difference between analog gain ISO and software ISO in order to get the best from our cameras.
 

[BJL]

I already agreed that there is some difference, what I am debating partly is the prejudicial use of "real" and "fake" for the two alternatives. Do not pretend that the word choice is arbitrary: "fake" has an obvious derogatory meaning! Would you mind if someone started using "real" only for digital amplification and "fake" for ISO adjustments done with analog gain? People have indeed claimed that analog gain is a fake change in ISO speed, and that only base ISO is a real ISO!

Your extreme example seems to consist of exposing at exposure index 1600 but using analog amplification to four stops less (ISO 100 setting), so pushing the analog signal strength down so low that it then suffers horribly from noise later in the processing pipeline, including probably quantization noise from the ADC. Which only shows that a fixed analog gain approach would probably use far higher than "ISO 100 level" with that sensor. Maybe closer to its "film equivalent (noise based) ISO speed", which is probably about 800 to 1600.

[Guillermo Luijk]

I already agreed that there is some difference, what I am debating partly is the prejudicial use of "real" and "fake" for the two alternatives.

There is not some difference, there is a big difference as seen in the examples above. I understand your point about the wording, but I don't consider the discusion about semantics too interesting, sorry. You are free to propose new names for the 'real' and 'fake' ISOs.

Your extreme example.

The example is not extreme at all. It is a part of an article explaining that in low light conditions (a very common situation), pushing camera ISO is very convenient vs digital amplification.

[peter.s.]

I didn't get this image:


If Npost symbolizes the digital amplification (aka fake ISO) then it should be placed after the A/D converter.

I agree that fake and real iso is a little misleading though. Real iso is not real anyway because the sensitivity of the sensor is not changed, only the amplification in the analog stage. Fake iso is definitely fake though because it's just a multiplication of raw values, an amplification in the digital domain. For a raw shooter there is no point in using fake iso settings because it will only cut headroom in the raw file without any benefits. The same result can be obtained by increasing EV in the raw converter but without the side effects. However for jpeg shooters it's needed because the images would look underexposed otherwise.

Maybe we should just call it analog gain and digital gain and be done with it. They use the terminology gain in video don't they? Gain is certainly used in the audio field and digital audio has been around longer than digital imaging.

[Guillermo Luijk]

If Npost symbolizes the digital amplification (aka fake ISO) then it should be placed after the A/D converter.

No, it symbolyzes (analog) noise introduced after the ISO amplification stage, i.e. at any point of the A/D conversion before the digital encoding process took place.

[peter.s.]

No, it symbolyzes (analog) noise introduced after the ISO amplification stage, i.e. at any point of the A/D conversion before the digital encoding process took place.

Thanks, that makes sense - N for noise introduced at that stage.

So adding to your original drawing it would look something like below if we add the additional digital ISO as well:

[Anders_HK]

I seem to recall seeing photos, such as those of space shuttle launches, where the ISO of the film used was 25. No, that's not a mistake, ISO 25. Why? Because the takeoff is a very very bright event, when you're close up. But doesn't a really fast shutter make up for that? In this case, yes, but not all...

Recently I was sitting on the rocks of a stream, during daylight hours, trying to get the "smooth water" effect. But even at ISO 100, using a high grade polarising lens, it's just too bright. With a bit of research, it seems that the trick these days is to use a rather dark filter. This cost of this approach is a(nother) bit of glass in front of the lens, potentially having a negative impact on your image quality (think of Reichman's notes about how the UV/haze filter of today really doesn't add any value and that it's like shooting through a window.)

Of course, the f-stop can be reduced down to, say, f/22 or smaller, but this too can have an impact on image quality due to diffraction and airy discs.

This all got me thinking and wondering if in the pursuit of high ISO perfection in low light, we're hurting ourselves at the other end and we just haven't realised this. For instance, various DSLRs now have a minimum ISO of 200. Those "smooth water" shorts are now going to require more effort than before this.

Of course ISO 25 is not going to be of benefit to those great big white lens folks that walk around sporting events nor the paparazi flooding a star with flashes at an interview. Nor is it going to help parents at their children's party or sporting event on the weekend. But now that the "film" is part of the camera body, for those that became used to low ISO with landscape photography, it seems like that choice has been removed. And whilst some cameras today are sold with ISO 50 capability, analysis seems to suggest that this is just  a corrected, underexposed photo at ISO 100.

Comments?

Hi,

Your post sure sparked replies advocating digital... at high (base?) iso. What they did not mention was that a higher iso also mean a more contrasty photo (lower quality digital capture, not to mention higher noise). Sure, advancement in digital techniques have led to better and better sensors, albeit some folks for optimum quality at least in certain circumstances still shoot FILM (slide or BW). I shoot MFDB with base ISO 50, and yes there is also no AA over sensor to degrade the image. You are correct, ANY filter over lens will also degrade quality, while indeed a lower ISO result in better image quality (as in for digital the base ISO, not extended). Some pros still use the 25 ISO in Leaf Aptus 22 MFDB, why? I also shoot film, and yes... exclusively Velvia rated at 50 ISO. Why? Clean files, and best image quality for landscapes in certain circumstances. At most I shoot 400 ISO and those times on my MFDB (but for landscapes no higher than 100 ISO). For quality images I do not see that one need more. Now... if we speak of sports or news photography where point is to nail a shot in dim light or of quick action just to sell, and thus with no more image quality than can be printed on web or in a newspaper... then it really does not matter of image quality beyond that requirement. Or... if point frank is to simply sell more new cameras that makes people upgrade and spend $$ every other year, thus making people believe that high ISO is the best thing since sliced bread can perhaps help... (!) Though from my eye, you are right. Low ISO still yields best image quality, but a tad forgotten among all else. Look around at big advertisement prints... many not even sharp. Why?

I am sure there will be posts arguing opposites to my above comments... but I frank write what I humbly see.

Attached, one shot digital, one shot film. I believe most can tell which is which. Digital tends to appear as to record color, while a quality slide film tend to record light (and with a magic interpretation of it). And yes, the attached are low ISO. For portrait I prefer digital hands down, for landscape, I am split. For blurring water... of course clearly LOWest iso (= 50).  

Regards
Anders

[Guillermo Luijk]

The reason to distinguish Npre and Npost was just to put into straightforward equations the explanation of why output SNR improves when pushing ISO that Emil Martinec pointed somewhere.

The total signal output from the ideal ISO amplifier is:

Samp = (S + Npre) * Giso

The total signal output from the ideal ADC is then:

Sad = (S + Npre) * Giso + Npost

The SNR ouput from the ADC stage would then be:

SNR = (S * Giso) / (Npre * Giso + Npost)

When pushing ISO (higher ISO gain, Giso), Npost becomes << Npre * Giso, so the SNR maximises to the limit:

SNRmax = S / Npre

Cameras where the Npost effect is higher than Npre (Canons) benefit more from pushing ISO. In cameras where Npre is higher than Npost (Nikons) the convenience of pushing ISO in low light conditions is not so critical.

Regards