Invariance

[Bill Pierce]

I’ve been asked to explain ISO invariance, essentially, “What is it?” At its simplest level it’s when the film speed settings on a digital camera are relatively meaningless in that underexposing at the base ISO setting of a camera and brightening the underexposed image in a processing program like Lightroom or Photoshop doesn’t produce a significant difference from exposing the image at a higher ISO and not brightening it in the processing program. That means no significant difference in shadow detail or noise level. In other words, ISO settings on the camera become relatively meaningless.

Some camera sensors do this; others don’t. The recent Fujis are ISO invariant. I’m told that the Sony A7RII, the Nikon D810 and D750 are. I’m sure there are other cameras that fall into the invariant category. It’s certainly easy to run a test on your camera to find out. And it’s worthwhile. Why? Because the big exposure problem with digital is blowing out the highlights. Once those highlights are overexposed, they are gone forever. Underexposing to make sure that doesn’t happen and brightening in “post production” can eliminate that problem. You would have to shoot raw; it’s not going to work with jpegs. You may have to change your camera set up to maintain viewfinder brightness. But it is definitely a technique worth exploring.

The mantra used to be “expose to the right.” It meant make sure your histogram moved fully to the right. In the early days of digital, this minimized noise which could be a real problem. I think we’re moving into the “expose to the left” period if your camera sensor is ISO invariant. There’s much to be explored here. Your thoughts?

 

[Ko.Fe.]

My Canon 5D was "expose to the left" and I remember some other users of this model also noticed.

 

[Pioneer]

My 5D is certainly that way. I shot an entire series of photos at the San Diego zoo several years ago after just reading about expose to the right. I put a lot of effort on that trip into ensuring that my photographs were all as close to the right as I could get without hitting the edge.

Everyone where the sky was involved, or other lighting, the highlights were well and truly blown.

I learned my lesson on that trip and have never tried to "expose to the right" with any of my digital cameras. For some odd reason I get pretty decent results.

Maybe "expose to the right" was always a poor technique in the real world? Maybe we will find out with better and better post processing software that most sensors are invariant to ISO.

 

[shimokita]

good point...

I have done a series of test shots and depending on one's ability with image processing software, comparable results can be obtained with raw files that are 'reasonably' exposed. Digital ISO, that is the ability to capture clean, has improved over the years... like scanning, the approach is to grab as much detail as possible. The technology behind digital photography, unlike the film industry, does evolve ; )

 

[Pioneer]

...The technology behind digital photography, unlike the film industry, does evolve ; )

I am not so sure that is completely correct shimokita. While digital has enjoyed the lion's share of attention...and money...film has also progressed.

Newer films are easier to scan, yet can still be effectively and wonderfully printed in the darkroom. New films continue to be produced and old films are improved. It is not happening at quite the same pace, but it still happens.

I think what is most interesting about film today is that it seems comfortable in both the analogue world where it was created, and in the digital world through scanning. Just about any old film scanner will work and the prices for these devices are pretty low.

Digital, for all of its improvements, is not as flexible. While it is possible to produce a digital negative and print using analogue methods, it still take unique printers and special ink sets for this to happen.

 

[user237428934]

If there was a real iso invariant system, there was no need to chose the iso in camera but that setting is still so prominent. And if there were real iso invariant systems, it wouldn't matter if I expose 1/1000 when the image really needs 30s exposure. The correction that is possible is rather limited, adjusting 4-5 EV values and that's it. In addition most people want to see the correct exposed image in advance in their EVF or Display or in the preview image after the shot. The camera needs a proper iso value for that. And there are a lot of JPG shooters. They can't adjust later in post processing.
Iso invariance is a characteristic of some sensors that works within limitations. That is the reason why no one advertises this feature, it's a feature that can be useful for a small group of people.

 

[michaelwj]

If there was a real iso invariant system, there was no need to chose the iso in camera but that setting is still so prominent. And if there were real iso invariant systems, it wouldn't matter if I expose 1/1000 when the image really needs 30s exposure. The correction that is possible is rather limited, adjusting 4-5 EV values and that's it. In addition most people want to see the correct exposed image in advance in their EVF or Display or in the preview image after the shot. The camera needs a proper iso value for that. And there are a lot of JPG shooters. They can't adjust later in post processing.
Iso invariance is a characteristic of some sensors that works within limitations. That is the reason why no one advertises this feature, it's a feature that can be useful for a small group of people.

This (exposing for 1/1000 vs 30s) is NOT ISO invariance - you are exposing the sensor to vastly different amounts of light.
ISO invariance is when the noise created by increasing the ISO is the same as pushing the exposure in post. So a shot at 1/250th, f/8, ISO 3200 has the same noise as a photo taken at 1/250th, f/8, ISO 200, then pushed 4 stops. The same number of photons are collected in each case, we've just changed where we added the gain. This may seem trivial, "why wasn't it always like this", but it really is a breakthrough in sensor technology.

 

[Ronald M]

My Leica M8 & 9 are definitely invariant. Neither will win any prize for high ISO. I really don`t care as my old favorite film was Kodachrome 25.

If I had to use them at high ISO, the system I worked out is expose at base 160 and push brightness in photoshop. The secret is discovering base is not noise free so when amplified the noise goes up also. I clean the 160 noise in Photoshop and then push. The higher I go, the better the results compared to "pushing" in camera and the less invariant it becomes.

Once I figured this out, I went on to D750 Nikon. There is improvement at 6000, but it is nowhere near as dramatic.

Why would I shoot D800 at high ISO? Camera was purchased for high quality. 100 is where mine are set.

Today I would buy a D500 if it were full frame. No desire to shoot crop any more.

As an aside , I always thought ETTR was not the best way to work.

 

[Timmyjoe]

Bill, I read through your explanation a number of times. If I'm understanding correctly, with ISO invariant cameras, changing the ISO from base to a higher number isn't making the camera sensor more light sensitive, it's just applying a brightening algorithm to the image after it is captured, to make it look like the camera sensor is more light sensitive?

 

[willie_901]

If there was a real iso invariant system...

Nikon sold at least one "real ISO-invariant" camera. The D7000 apples only digital multiplication to increase post-acquisition brightness. There may be others.

Many cameras approach ISO invariance which means the increase in noise (except for photon noise) as ISO increases is negligible. Other cameras can have at least one range of ISO parameters that are strictly ISO invariant. The only way to evaluate the degree of ISO invariance is statistical analyses of un-rendered raw files' noise levels.

I guess an ad-hoc definition of an ISO-inavriane is when you pixel peep and can't tell the the difference between shadow-region signal-to-noise ratios in images where brightness at base ISO is achieved during post-production or set in-camera using electronic amplification of the signal before it's digitized.

For cameras with OVFs seeing an appropriately bright scene in the finder is a given. With EVFs many other brands let you choose between seeing an approximation of the brightness determined by shutter time and aperture, or an image that automatically brightened based on the light meter's estimate. This means exposure techniques that take advantage of ISO invariance are problematic in two situations: for JPEG shooters, or when circumstances require in-camera image review.

 

[Scapevision]

Maybe "expose to the right" was always a poor technique in the real world? Maybe we will find out with better and better post processing software that most sensors are invariant to ISO.

Some of the early XXD and XXXD series cameras really benefited from exposing to the right. It saved me many times shooting landscapes with the non-landscape cameras 😀

 

[willie_901]

Bill, I read through your explanation a number of times. If I'm understanding correctly, with ISO invariant cameras, changing the ISO from base to a higher number isn't making the camera sensor more light sensitive, it's just applying a brightening algorithm to the image after it is captured, to make it look like the camera sensor is more light sensitive?

Increasing ISO almost never makes the sensor more sensitive[1]. The sensitivity is fixed at the camera's base ISO.

As mentioned by others in this thread, ISO-invariance is when electronic amplification of the signals after the shutter closes does not meaningfully increase the electronic noise levels.

[1] Today there are three APS-C cameras (X-Pro2, XT-2 and a6000) using a SONY sensor with dual-gain technology. These cameras have two base ISOs (200 and 800)(link). No doubt this technology will become more common.

 

[user237428934]

As an aside , I always thought ETTR was not the best way to work.

If you use it correctly you make sure that you will never lose any highlights. But sometimes you shift the historgram with ETTR so much to the left that overall image quality suffers. For example if you have lot's of darker areas and you cramp them together on the left side and try to recover the dark areas again this can cause lots of grain.
Problem ist, that histograms of most cameras show a jpg-histogram and the computer in your camera has no idea about how capable your raw-converter in restoring highlights is.
Try this: take a bright scene and find the exposure so that your histogram does not show any blown highlights. Then take several shots adding 1/3 EV after every shot. Restore the highlights with your raw-converter and find out, how much latitude you have.

I bet a lot of people who complain about the grain when rescuing shadow details with Canon sensors used ETTR which brought them at least 1EV of problems in the shadows.

 

[willie_901]

Expose To The Left OR The Right.

Expose To The Left OR The Right.

Exposing either to the left or the right can produce images with the highest possible signal-to-noise ratio. It depends on the situation at hand.

Rather than thinking about exposing to the left or right, with ISO-invariant cameras, consider Prof. Emil Martinec's advice to maximize exposure.

"What is the appropriate mantra?* I would prefer "Maximize Exposure"; maximize subject to three constraints:

(1) maintaining needed DoF, which limits how much you can open up the aperture;
(2) freezing motion, which limits the exposure time;
(3) retaining highlight detail, by not clipping wanted highlight areas in any channel.*

Note that ISO is not part of exposure.* Exposure has only to do with aperture and shutter speed.* Maximizing exposure guarantees that one captures as many photons as possible subject to photographic constraints, and therefore optimizes S/N.


Prof. Martinec presents a technical discussion on the physics of digital imaging here.

 

[Godfrey]

The question of "ISO invariance" is a matter of how a particular camera/sensor achieves the sensitivity settings.

1- All imager sensors have a base sensitivity—a natural reaction to light falling on them: X amount of light input over N units of time equals Y amount of electrical output. That output goes to the A->D converter subsystem for filtering, amplification, and reduction to a discrete numeric representation.

Base sensitivity is achieved at the design nominal settings for the sensor with respect to its overall voltage environment. Sensitivity can be altered at the input to the A->D converter chain by altering the voltage environment, with some losses (either increased noise or decreased dynamic range). This sort of sensitivity adjustment is relatively fragile and not often done.

2- Amplification to adjust sensitivity can happen in the A->D converter chain, this is most commonly where hardware-based sensitivity adjustment occurs. Just as with pre-input adjustment, some losses occur here.

3- Digital adjustment (or amplification) of ISO—that is, multiplication of the output signal in the digital domain—happens after the A->D converter chain. This can be very low loss if the sensor has enough dynamic range and low enough noise at base ISO. The signal can be filtered during the multiplication process as well to remove excess noise and prevent loss of dynamic range by using a non-linear multiplication map.

Most cameras achieve ISO adjustments with a combination of #2 and #3 techniques. What constitutes ISO invariance is when #2 is used for a very limited range of settings and the rest of the range is achieved with #3 exclusively. Then, the retention of dynamic range and noise control between setting the ISO in the camera vs in the post-processing environment depends on the quality of the digital amplification provided by both, and where in the raw conversion process the amplification by the post-processing software is applied.

With the cameras I've had, ISO invariance was most easily observed with the Kodak CCD sensors of the 2003 Olympus E-1 and 2009 Leica M9.

The Olympus E-1 stepped ISO adjustment in hardware from ISO 100 to ISO 800, then went to a purely digital amplification for ISO 1600 and ISO 3200. Because of the development in raw conversion algorithms between when it was released and today, while its original performance was somewhat poor at ISO 800 and above, modern raw converters applied to E-1 raw files now give remarkably good performance even at ISO 3200. But because of the relatively high-ISO at which pure digital amplification was incorporated, most of the "ISO invariance" happens in the upper three ISO settings.

The Leica M9 stepped ISO adjustment in hardware from ISO 100 to ISO 640, but seems to have a tapered hardware step through the range—it gains ISO more and more by digital amplification than by hardware amplification through the range. This much more recent CCD sensor also has significantly improved dynamic range and lower noise than the older E-1 sensor. It's behavior when used with an ISO invariant exposure technique is such that exposures made at ISO 400 or 640, and amplified in post processing to ISO 3200, show almost no difference at all from exposures made at ISO 3200 setting, and can be quite clean and DR invariant depending on the raw converter used.

The downside of exploiting ISO invariance exposure technique, of course, is that the recorded image that you can view on the camera's LCD will be very dark and impossible to judge focusing accuracy or shadow values with. So exploiting ISO invariance basically disables two elements of digital capture that many photographers use to good effect: a live or just-post-exposure view of their exposure judgement, and the ability to ascertain critical focus in immediate review.

My own philosophy of digital exposure is simple: exposure that nets the effects you are after and returns a full range of data makes for the best image rendering capabilities. It's quite similar to my philosophy of film exposure: exposure + processing that captures the range from shadows to highlights that I'm after nets the best negative for printing (or transparency for projection). I use whatever characteristics of either capture medium I find, through testing, to achieve the best notions of exposure evaluation to achieve those best exposure points. Simplistic exposure methodologies like "expose to the right" are never more than a rule-of-thumb or first order approximation.

G

 

[willie_901]

Here is a 2015 presentation titled "ISO-less?" by Prof. Hank Dietz & Paul Eberhart, University of Kentucky Electrical & Computer Engineering.

While this is a technical presentation some of their explanations, results, ideas and proposals are interesting.