Any advantage to higher ISO when shooting in RAW?

[spacediver]

My (basic) understanding of ISO settings on a camera is that it increases the gain of the signal during the amplification of stored charge into the voltage signal (which is then converted to digital).

In terms of information, is anything gained by increasing ISO?

I can see how it might a situation where two objects in the scene differ in brightness so minutely that even though the sensor physically captures this difference, the resulting voltage difference for the pixels representing these two objects isn't great enough to be encoded as a different pixel value (although if you could arbitrarily increase the bit depth of the RAW file this wouldn't be an issue).

A sharper way of asking this question is as follows:

Suppose you had a camera that could encode pixel values with as many bits as needed (i.e. you could differentially encode, into pixel values, any physically detected difference). Also assume you're shooting RAW so you have full control over the assignment of pixel values to rendered tonal values.

Would there then be any benefit (other than convenience) to shooting with anything but the lowest ISO setting?

 

[robbins.photo]

My (basic) understanding of ISO settings on a camera is that it increases the gain of the signal during the amplification of stored charge into the voltage signal (which is then converted to digital).

In terms of information, is anything gained by increasing ISO?

I can see how it might a situation where two objects in the scene differ in brightness so minutely that even though the sensor physically captures this difference, the resulting voltage difference for the pixels representing these two objects isn't great enough to be encoded as a different pixel value (although if you could arbitrarily increase the bit depth of the RAW file this wouldn't be an issue).

A sharper way of asking this question is as follows:

Suppose you had a camera that could encode pixel values with as many bits as needed (i.e. you could differentially encode, into pixel values, any physically detected difference). Also assume you're shooting RAW so you have full control over the assignment of pixel values to rendered tonal values.

Would there then be any benefit (other than convenience) to shooting with anything but the lowest ISO setting?

Is there a benefit to higher ISO? Absolutely - it allows you to either stop down the lens, increase the shutter speed or both and get the same exposure that you would at a lower ISO. So depending on the conditions there is a definite benefit to using a higher ISO.

 

[Rick50]

One of the most useful applications is when you reach the 30 second shutter speed camera limit. By kicking up the ISO you get the exposure time back inside 30 seconds. I also often use Auto ISO when hand holding indoors. I set the shutter speed to 1/focal length in Shutter priority and let the camera adjust the ISO when it can no longer open up the aperture any more.

 

[Gary A.]

In "terms of information" ... yes, increasing the ISO delivers information/data where at a lower ISO there wouldn't be any information/data thereby and ultimately rendering a detailess black.

 

[spacediver]

In "terms of information" ... yes, increasing the ISO delivers information/data where at a lower ISO there wouldn't be any information/data thereby and ultimately rendering a detailess black.

would this statement hold true even in my imaginary scenario where you could encode an arbitrarily small increase in voltage with a change in pixel value (i.e. infinite bit depth)?

 

[Gary A.]

In "terms of information" ... yes, increasing the ISO delivers information/data where at a lower ISO there wouldn't be any information/data thereby and ultimately rendering a detailess black.

would this statement hold true even in my imaginary scenario where you could encode an arbitrarily small increase in voltage with a change in pixel value (i.e. infinite bit depth)?

I'm not an engineer or theoretical mathematician ... but I get hung up on "arbitrarily" as all values need a baseline and then need to feed off one another. Arbitrarily seems to directly conflict with infinite bit depth ... if you have unlimited bit depth I suspect you wouldn't need to elevate ISO and shift your dynamic range, the entire spectrum would have detail.

 

[spacediver]

thanks, yea it seems to me that increasing ISO compresses the available dynamic range into whatever available bits there are. Which I can definitely see being useful in low light conditions where you're interested in "zooming" into the low light range.

 

[bratkinson]

ISO speed is one of the three basic parts of the exposure triangle. Adjusting ISO speed is one of the means available to obtaining the desired exposure. Low light photography demands a combination of fast lenses (f2.8 and wider, typically), slower exposures, and higher ISO speeds.

But like all aspects of the exposure triangle, changing any of them has both positive and negative effects. As ISO speed is increased, more noise will result. I'm not an electrical engineer, but from what I understand, the higher ISO speeds force the camera circuitry to 'interpret' ever lesser voltage outputs from the sensels more 'loosely'.

It's a throwback to the early days of computers...with which I am quite familiar. One of the first generation computers I worked with in 1966 was an IBM 1620. This was 'unique' in that it was a DIGITAL computer rather than a BINARY computer, like those before and since. Using voltage values, the IBM 1620 would treat 0.51 volts to 1.50 volts as a '1'. and 1.51-2.50 as a '2', etc. But as the individual circuits got 'older' (tubes, what else?), what used to be put out as, say, 1.55 volts dropped to 1.49...OOPS!!! Wrong number!

In my mind, raising ISO speeds too fast gets similar results...after ISO speed-based 'amplification' of 0.06 millivolt to 0.54 millivolt (I'm guessing wildly, here) does 0.54 get treated as an 'ON' or 'OFF' bit? How RELIABLE is that 0.06 millivolt to begin with? Then amplify it by a factor of 9? To look at it scientifically, the -reliability- of such a result is likely in the range of plus or minus 30-40%. One's odds playing blackjack at a casino are far better than 30-40%!

Obviously, the newer digital cameras handle high ISO speeds better than those from before. As I do most of my photography at indoor church events as well as nighttime outside, I upgraded several times before ultimately getting a Canon 5Diii, primarily for its high ISO capabilities and its dead-on auto-focusing.

 

[Big Mike]

Either you're way over-thinking this, or you're talking about stuff that is way over my pay grade.

Your 'basic understanding' of ISO is right in line with how I understand it. But the main way that I apply that understanding, is to consider the signal to noise (s/n) ratio.

The amount of noise that is generated for a given ISO level, is pretty constant. But what isn't constant, is the amount of signal (which of course, is derived from the brightness/exposure of that part of the picture). And it's that s/n ratio that will determine how much noise you actually see in the image. This is why you can sometimes shoot at lower ISO and see more noise, and sometimes shoot at higher ISO and see less noise (or at least an acceptable amount considering the higher ISO).

This is why I employ the 'expose to the right' technique. When you expose your image to a brighter value, the sensor creates more signal (more information), this will improve your s/n ratio and thus the noise won't be seen as much in the image.

When I try to teach this to a group of students, it sometimes doesn't sink in very well, so I simplify it by saying that you'll get less noise by slightly overexposing and 'normalizing' the image a bit darker in post. And you'll get more noise if you underexpose and have to increase exposure in post. So if your histogram shows room for increasing exposure without clipping, then that would be advantageous, but we should also take care not to underexpose images, because fixing them will really start to make them noisy.

 

[Ysarex]

In terms of information, is anything gained by increasing ISO?

No.

The light sensitivity of the sensor is a fixed physical property of the sensor and is not variably adjustable. You expose the sensor and it records information within the threshold limits of it's exposure range. If you expose the sensor so that it's maximum capacity is used you get the most information. If you expose the sensor less than it's maximum capacity you get less information. Amplification of the signal (ISO increase) occurs after exposure and the amount of recorded information at that point is already fixed. If you have less information and you amplify it you still have less information. Amplification doesn't add any information that you didn't get in the first place.

Joe

 

[Vince.1551]

No added advantage shooting at raw. Application of signals degrade IQ. Best to avoid high ISO when possible.


Sent from my iPhone using Tapatalk

 

[photoguy99]

With infinite bit depth, there would be no benefit to increased ISO. With infinite bit depth you can amplify in the digital domain just as well as in the analog domain.

Since bit depth is not infinite, it makes a difference whether you amplify in the analog domain with increased ISO or in digital domain with RAW processing. The differences are surprisingly subtle. Ysarex showed some very interesting sales in a similar thread not long ago, and we had a nice discussion about what kinds of pros and cons there are.

 

[spacediver]

I'm not an electrical engineer, but from what I understand, the higher ISO speeds force the camera circuitry to 'interpret' ever lesser voltage outputs from the sensels more 'loosely'.

Yep, this is my interpretation too. Can be useful when working in low light conditions where there's a higher likelihood that those ever lesser voltage differences actually correspond to physical differences within the scene that you are interested in capturing.

 

[spacediver]

The amount of noise that is generated for a given ISO level, is pretty constant. But what isn't constant, is the amount of signal (which of course, is derived from the brightness/exposure of that part of the picture). And it's that s/n ratio that will determine how much noise you actually see in the image. This is why you can sometimes shoot at lower ISO and see more noise, and sometimes shoot at higher ISO and see less noise (or at least an acceptable amount considering the higher ISO).

When I try to teach this to a group of students, it sometimes doesn't sink in very well, so I simplify it by saying that you'll get less noise by slightly overexposing and 'normalizing' the image a bit darker in post. And you'll get more noise if you underexpose and have to increase exposure in post. So if your histogram shows room for increasing exposure without clipping, then that would be advantageous, but we should also take care not to underexpose images, because fixing them will really start to make them noisy.

Thanks, that makes quite a bit of sense. Just to clarify, would it be more accurate to say that lower ISO always produces less noise than higher ISO, but the signal to noise ratio may sometimes be worse at lower ISO?

 

[spacediver]

No.

The light sensitivity of the sensor is a fixed physical property of the sensor and is not variably adjustable. You expose the sensor and it records information within the threshold limits of it's exposure range. If you expose the sensor so that it's maximum capacity is used you get the most information. If you expose the sensor less than it's maximum capacity you get less information. Amplification of the signal (ISO increase) occurs after exposure and the amount of recorded information at that point is already fixed. If you have less information and you amplify it you still have less information. Amplification doesn't add any information that you didn't get in the first place.

Joe

Amplification may not add information, but the partitioning of the voltage signal into bits that represent information in the scene may have an impact.

For example, suppose the scene consists of four dark patches with very subtle differences in lightness.

Suppose your camera has a 2 bit analog to digital converter, and can thus only encode four different values per pixel.

Also assume 100% quantum efficiency of the sensor and zero noise.

Take two scenarios:

1: Lowest ISO
2: Higher ISO

In both scenarios, the sensor will physically detect differences between the four patches, but in the first scenario, the four patches may not be differentially encoded into the pixel values (i.e. the pixels would all have the same value).

In the second scenario, there might be an ISO that is just high enough that the physically detected differences between the four patches map perfectly onto the four different ranges of voltage level that correspond to the available bits. In this case, all four patches will be differentially encoded, and there will be more information (in the pixel domain).