The Effect of High ISO on Camera Image Resolution

Image resolution is generally considered to be unaffected by the ISO being used. This assumption has always left me feeling slightly uncomfortable. I have always used ‘normal’ ISO ranges when evaluating lens resolution (typically between 100 and 400), because it doesn’t make sense to do an analysis outside of typical shooting conditions.

I have had a suspicion that extremely high ISO’s would eventually cause a plunge in measured resolution, and what follows are some typical results from those tests. Spoiler: I didn’t get what I expected to get.

I’m not referring to lens resolution in my tests, because resolution measurements are a combination of camera sensor characteristics, the lens, lighting, and (possibly?) the ISO in use.

Years ago, I had done some tests with resolution-versus-ISO, but I stopped at ISO 6400, since my camera at that time had severely degraded image quality even at ISO 3200. Through that modest ISO range, there wasn’t much of a change in measured resolution. Modern sensors are really, really good, and when I use Topaz Photo AI for noise reduction I get impeccable results at high ISO’s.

I did the following tests using my Nikon Z8 with the Meike 85mm f/1.8 lens, shooting at f/2.8. I use the MTFMapper program to measure the resolution, and my resolution target is 42”X57”, or 107cm X 145cm. I used a wired shutter release, and my mirrorless camera doesn’t suffer from mirror or shutter vibrations to affect sharpness even at low shutter speeds. I also keep IBIS active, which works well even on a tripod.

After I focused the lens on the resolution target, I took each (raw) photo after only adjusting the ISO value, and never re-focused. For the measurements, I used the unedited raw shots.

Resolution target, ISO 400 (left) and ISO 51200 (right)

Sample shots of the annotated resolution target are shown above after processing in the MTFMapper program. These are two shots out of the 11 shots taken, ranging from ISO 100 through 51200. The little trapezoids in the target get measured on all 4 sides, with the MTF50 measurement overlaid on the edge, displayed in “line pairs per millimeter” units.

Resolution plots, ISO 400 (left) and ISO 51200 (right)

Resolution plots were made for each shot. You would think that the plots would look more similar than they do, since the focus wasn’t changed in any of the shots. I had expected to see a plunge in the resolution at the extreme ISO 51200 shot, but that isn’t what happened. I did a series of tests, where I refocused before each series, and each time I got similar measurement results and trends to what is shown above.

Zoom in on same target location, ISO 400 (left) and ISO 51200 (right)

The left side shows ISO 400 measurements at the same location in the target as the right side, which was shot at ISO 51200. The ISO 400 shot got an MTF50 measurement of 74.6 lp/mm on the same edge that got a measurement of 85.9 lp/mm in the ISO 51200 shot. The ISO 51200 shot looks darker, but theoretically that should reduce contrast/resolution instead of increasing it. So much for my theory that the measured resolution would plunge at extremely high ISO’s!

The high ISO shots start getting pretty noisy, but that doesn’t seem to be a problem in regards to measured resolution.

For this lens, the sagittal (wheel spoke) direction is much sharper than the meridional direction when you get away from the lens center.

At this point, I need to mention the “Nyquist Limit”, which limits the camera sensor resolution according to the pixel density. For my Nikon Z8/Z9 sensors, this limit is (8280pixels / 35.9mm)/2 or 115.5 lp/mm. Another factor, called the “Kell Factor”, takes into account the space between the light-sensitive portions of the pixels. This Kell factor is 0.7, so the practical sensor resolution is (115.5 * 0.7) or 80.85 lp/mm. Any resolution measurements higher than 80.85 lp/mm are considered unreliable for this sensor.

Web sites that quote resolution measurements beyond 80.85 lp/mm for a lens measured using a camera sensor equivalent to the Nikon Z8/Z9 should be considered suspect.

Peak MTF50 measurement at each ISO

I noted the highest MTF50 measurement in each shot of the target, and plotted the result. There was a clear pattern of increased resolution as ISO was increased, which is exactly opposite of what I would have expected. At lower ISO’s, the resolution change is actually pretty minimal as the ISO changes. As ISO increases, the resolution seems to climb at an accelerating pace instead of falling off of a cliff.

Now, it’s time to step back and do a reality check. Programs that measure resolution do so by looking at the number of pixels on the sensor required to go from the target white background to the targets’ fully-black edge. But is that really the best measure of resolution?

Seimens Star, ISO 51200 (left) ISO 400 (right)

I took some shots at 7.7 meters away with my 85mm lens at f/2.8 with ISO 51,200 and ISO 400. What’s shown above is the cropped portion of the photos with the Seimens Star target. This is what ‘real’ resolution looks like. Indeed, you’d have to conclude that high ISO in actual fact does ruin resolution.

Current resolution-measuring software (like Imatest and MTFMapper) don’t know how to evaluate closely-spaced lines to estimate resolution, and instead depend upon measuring the pixel distance between light and dark on an edge. This ISO test shows that the edge contrast-measuring technique for determining resolution can fail spectacularly when considering high ISO.

Web sites rarely specify the ISO they use when measuring lens resolution, but it looks like there’s yet another way to manipulate the results. Shooting with jpeg instead of raw format is probably the biggest fraud to artificially create better lens resolution measurements, but using a really high ISO is evidently another way.

ISO 51,200 after using Topaz Photo

Just for fun, I tried using Topaz Photo on the ISO 51,200 shot shown above. Still hopeless.

ISO 500 versus ISO 102400, 100% crop at 15 meters

For an even more extreme example, I shot a label with barcodes on it, as shown above, from 15 meters. This is an un-sharpened photo with the pixels shown at 100% magnification. You can still read the ISO 500 text and see the bars, but the ISO 102400 shot is pure garbage. Pretty outrageous that resolution-measuring software is actually happier with the 102400 shot.

At extreme ISOs, the image contrast can get quite high. This high contrast is what fools the resolution software into concluding that the resolution is high, when it’s obvious that the resolution is actually terrible.

If you want the best resolution, definitely stick with low ISO (even if the resolution-measurement software doesn’t think you should).