Is it better to shoot with a higher ISO, or use lower ISO and raise the exposure in post-processing?

  • Suppose there are two photos taken in the following scenario:

    Scenario 1 - Photo taken using ISO of 3200.

    Scenario 2 - Photo taken using ISO of 100.

    Both photos are taken in a medium light scenario. The photo in Scenario 2 is quite dark, while the photo taken in Scenario 1 looks good.

    From my understanding, higher ISO gives a more grainy photo. So could a photo in scenario 2 (which appears quite dark) be modified in software (e.g. Photoshop) to increase the brightness, and be better than the photo taken with a higher ISO?

    I believe Scenario 1 will yield the better image with less noise - see here:

  • Matt Grum

    Matt Grum Correct answer

    8 years ago

    From my understanding, higher ISO gives a more grainy photo

    I'm afraid your understanding is incorrect. High ISO doesn't necessarily give a more grainy photo as there are other factors involved. In some circumstances it can be the case that lowering ISO increases noise. I did an experiment a while back to prove this:

    What you're seeing is exactly what you describe, one image shot at ISO 1600 unedited, versus the same photo shot at ISO 100, brightened in post to match the first image.

    As you can see the ISO 100 image is considerably noisier.

    What's going on is that images contain shot noise and read noise. Shot noise occurs because photons are emitted randomly by lightsources which gives rise to variations in the light hitting each part of the sensor. Read noise occurs as the analogue signal is transferred from the sensor to the ADC.

    What the ISO setting on the camera does is amplify the analogue signal before readout and digitization. When amplifying the signal the shot noise gets amplified too, thus the signal to noise ratio is the same. However the read noise does not get amplified as it happens after amplification. When you shoot at low ISO and brighten the image in software, the photon noise, and the read noise both get amplified, giving a higher level of noise.

    So why is ISO deemed to determine image noise?

    It is true that the lowest possible level of noise is achieved at the lowest possible ISO with as much light as possible hitting the sensor.

    It is also true that the greatest speed for my car is achieved in fifth gear. But this doesn't mean putting it into fifth with the engine idling will increase my speed. The biggest influence on speed in the throttle position. Gearing just prevents stalling or overrevving the engine.

    In the same way the biggest influence on noise is the total amount of light falling on the sensor. But you have to set the ISO to avoid over or under exposing the image.

    ISO does not have a direct relationship with noise, if you have a very dim scene with little light falling on the sensor then you will have noise regardless of the ISO setting. Likewise if you have tons of light falling on the sensor you will have very little noise regardless of the ISO setting (though you might have overexposure!)

    Another problem with perceptions of noise and ISO is that in any of the automatic shooting modes P/Tv/Av increasing the ISO will cause the camera to alter the shutter speed and/or aperture which results in less light hitting the sensor which means more noise. However in manual mode for a fixed shutter speed and aperture, increasing the ISO will not result in more noise being present in the image.

    So why does any of this matter? After all if you want the lowest noise your camera can offer you just set it on a tripod, pick ISO 100 and leave the shutter open as long as possible before the image becomes overexposed.

    The problem is that thinking about ISO first can lead to misunderstandings. For example when using the above approach in dim light with the aperture wide open, the shutter might max out at 1/30s. People remember the ISO 100 = lowest noise part, forget about the need to get a correct exposure (or are mislead by the image on the back of the LCD, which can look well exposed when you're viewing the screen in the dark) and end up underxposing the image, giving more noise than if they'd shot at say ISO 400.

    It is just as correct to say the lowest possible level of noise is achieved when as much light as possible is hitting the sensor (without overexposing the image) and the ISO is as high as possible (without overexposing the image). In most cases the highest ISO possible will be 100.

    Thinking about the level of light first, then the ISO avoids pitfalls when there is some limit to how much light you can get onto the sensor during the exposure.

    There are other minsunderstandings that originate from thinking about ISO as the primary factor that determines noise. One such misunderstanding relates to a camera's base (minimum native) ISO. Someone with a camera Y whose base ISO is 200 might think "since ISO 200 gets me the cleanest images, wouldn't it be great to have ISO 50 like camera X". Now it may be the case that camera Y has a sensor with fantastic quantum efficiency, and very good microlenses meaning it is very efficient at capturing light, hence images become overexposed quickly, leading to a high base ISO. Camera X might have a much older sensor with poor QE, no microlenses and low fill factor. It wastes a lot of light and thus requires longer exposures. It also produces images with the same level of noise at ISO 50 as Y does at ISO 200. Just because the number is lower doesn't mean it's better.

    Finally thinking about light first helps explain "ISO less" sensors, such as the latest batch of Sony sensors found in cameras like the Pentax K5 or Nikon D800. The read noise is so low that it doesn't make any difference if you amplify the signal prior to readout, meaning that you can get comparable results at many different ISO settings, proving that ISO isn't responsible for image noise.

    My preferred method of shooting is in manual mode with auto-ISO. This allows me to select the depth of field and amount of motion blur that I want/can tolerate in the image, and then have the camera minimise noise for me.

    Nothing like a simple test to disprove hours of speculation! +1! Also, now that I have auto-ISO in M, I hardly leave it. Greatest advancement in my recent body upgrade, along with silky ISO 1600 :)

    "The reason for the association between high ISO and noise..." Matt, that's not exactly correct, and you know it. If the image can be *correctly* exposed at a lower ISO (and let's assume an integral multiple of the camera's base ISO to get poor implementations out of the way), it will have both lower noise (resulting from a more statistically valid sampling of the light—a probabilistic phenomenon—or reduced shot noise if you prefer) and a higher dynamic range. In your example, a still life, the ISO 100 exposure could easily have been 1/2s, resulting in less noise than the ISO 1600 exposure.

    (cont'd) I'm not taking issue with the overall statement—that exposing correctly will be better than "fixing in post"—it's just that you keep leaving out the part where you should say that *correct exposures* at lower ISO values (at least until the base ISO is hit) will result in lower noise than *equally well-exposed* shots at higher ISOs.

    Stan has nailed it. You get the lowest apparent noise with the highest possible signal to noise ratio and pixel saturation. Sometimes you have no option but to select an aperture and shutter speed, and push ISO as high as you can without clipping a that aperture and shutter speed...but generally speaking using the lowest ISO setting while still getting a proper exposure will result in the cleanest images. If you have the ability to expose for several seconds, or use an ultra wide aperture, while pushing the exposure to the right, at ISO 100...then do so. You'll have a cleaner image.

    In the astrophotography world, people are using Canon DSLRs extensively and have split the hairs on the ISO question. Craig Stark has run the numbers on some Canon product and published a detailed analysis of linearity of data from these cameras: Profiling the Long-Exposure Performance of a Canon DSLR (July 2, 2012, Cloudy Nights);-Reviews/CanonLinearity.pdf In a nutshell, he says that Canon is modifying the dark signal before it hits the raw file and that settings above ISO 400 are of limited value.

    @jrista Max SNR is indeed achieved at min ISO with as much light as possible hitting the sensor. You say **"using the lowest ISO setting while still getting a proper exposure will result in the cleanest images"**, the problem I have with this is that it places the emphasis on ISO. The approach of "set ISO to 100 and then make sure the image is properly exposed" will guarantee the cleanest images **if there is enough light**, but what happens when the shutter maxes out at 1/30s is people remember the ISO=100 bit and forget the second part, leading to increased noise. I've seen it happen.

    ...continued... Whereas, approaching the problem from the other direction: **"get as much light as possible onto the sensor without clipping, then set the ISO to make sure the image is properly exposed"** is a methodology that guarantees the cleanest images **in all possible scenarios**. Thinking about amount of light first and then ISO avoids any trap, and prevents the type of confusion that resulted in this question being asked.

    @StanRogers I've rewritten that whole section, hopefully it makes more sense and is less ambiguous now.

    I am not saying always use ISO 100. I am saying push the histogram as far to the right as you can with shutter and aperture, then use the LOWEST ISO setting that gets the right-hand ends of the RGB histograms into the right-most histogram box. I guess whenever you post answers like this, it sounds like you are telling people to use the highest ISO setting possible, and I just want people to be clear...they should still be aiming to use the **lowest ISO** that maximizes the signal without clipping. If someone has the option of increasing exposure time or is ok with a faster aperture...

    ...using those options FIRST is better than increasing ISO. A wider aperture or a longer shutter are not always possible in all forms of photography, however there are many forms of photography where it is possible to increase shutter speed to maximize SNR. Your example with the books is an ideal is handy as a ready-made demonstration of auto-exposure in post ISO 100 vs. 1600...but you could have easily lengthened the exposure in that still-life scene and the ISO 100 image would have been better than the ISO 1600 image. You don't adequately explain that, and I think it is confusing.

    @smigol: As far as I can tell, that article is written rather specifically from the standpoint of astrophotography, and the elimination of dark current noise. Multi-minute exposures at higher ISO don't generally occur in any other form of most cases, the purpose of using a higher ISO is to achieve a **faster** shutter speed, meaning exposure times will usually be short, and therefor not encounter the long-exposure dark current issues. Broadly applying Craig's claims to all photography is invalid, as his context was set to astrophotography on the first page of the paper.

    Absolutely it's a specialized case. The important thing to note is that Canon is doing noise reduction at the raw, creating part of the different performance of the ISO levels. Thus, the example given may not be identical and it may not be as strictly controlled as originally thought. Still, this is a corner case.

    @smigol In that document, Craig has totally neglected readout noise. He suggests going beyond ISO 400 is worthless because discretisation error from counting single electrons is about the same as quantisation error from the A-D converter. However increasing ISO continues to be useful until discretisation error from electron counting is bigger than readout noise ( in terms of its standard deviation). From the bottom left figure on p3 of Craig's document it looks like the readout noise is about 10 A-D counts (sqrt of variance at 0 exposure) so ISO increases up to about 4000 would be useful.

    Hi, Matt! Thanks for your detailed explanation and post in this thread. Do you have an updated link to the tests you'd done? I'm getting a 404 when I click.

License under CC-BY-SA with attribution

Content dated before 7/24/2021 11:53 AM