Lens Resolution Measurement: Avoid Sharpened Jpeg Like the Plague
You can see wildly-varying lens resolution measurements for the exact same lens model out there on the internet. Do manufacturers really make lenses with that much variation? I think not.
Many (most) internet sites that show lens resolution measurement results don’t divulge how their measurements are done. Some sites actually state that they use jpegs of their resolution target straight out of the camera. Those same sites don’t tell you how much sharpening was used for those jpegs. What you do notice, however, is that they invariably show lens resolution results that are “too good to be true”.
The way you’re supposed to capture resolution chart images for analysis is with un-sharpened RAW. Only. And leave them that way. Exposure isn’t too critical here, but generally light meters will make black-and-white charts end up with whites looking too grey, unless you boost the exposure a little.
Why does it matter how resolution charts are photographed? Because resolution is based upon the transition from light-to-dark on target edges. Modern resolution measurement software is based upon how many pixels it takes to go from the white chart background to the maximum black on a target edge. The faster that transition occurs, the higher the resolution measurement you’re going to get. Details of the measurement process are discussed in this article.
How does sharpening of a photo work? By altering the light-to-dark transition on edges of objects in the photo. Do you see the connection? You can basically dial in the desired test results by adjusting the sharpening. Now your test results are meaningless.
Internet sites that provide lens resolution information should also discuss what kind of camera was used (assuming the measurements include the use of a camera sensor). The sensor resolution and whether or not the sensor has an “optical low-pass filter” (OLPF) is important information. An OLPF will lower the measurement numbers that get quoted.
If you don’t know this information, then you can’t compare one site’s lens measurements against another site’s measurements.
I think an example is in order, to prove the point. And because talk is cheap.
I am using the MTFMAPPER program, but programs like Imatest work the same way. They all find (slanted) edges in the photo, and count how many pixels it takes to go from white to black. When they know the size of the camera sensor pixels, how many pixels are in a row or column of your sensor, and how big your sensor is, then they can give you resolution measurements in a variety of different ways. You might get readings such as “cycles per pixel”, “line pairs per picture height”, “lines per picture height”, etc. at a particular contrast level (like 50%).
The resolution chart with lots of edges to measure
The chart shown above is a typical “slanted edge” resolution chart. You photograph it with the lens, camera, aperture, distance, and zoom setting you want to evaluate. Each edge of the little trapezoids will get measured by software to determine the lens resolution at that location in the field of view. For optimal results, the chart should just barely fill the field of view and be absolutely parallel to the camera sensor. The chart should also be parallel to the edge of the camera frame (for an optimal ‘slant’).
Resolution Measurement Comparisons
RAW, unsharpened chart: How it is supposed to be done
Shown above is the two-dimensional MTF50 chart plot, showing the “line pairs per millimeter” (lp/mm) measurements from the un-sharpened Raw photo of the test chart. This is a really good lens, and the peak measurements around 62 lp/mm indicate how good the lens is. This is the picture format of the resolution target that should be used for analysis.
Same chart shot, but now Jpeg sharpened in LightRoom. Amount = 36, Radius = 1.0
The moderately-sharpened jpeg of the same chart photo shows some too-good-to-be-true resolution measurements. Everybody would be standing in line to buy this baby, if these measurements were actually legitimate.
I used Lightroom to adjust the original .NEF raw photo with very modest sharpening, and then exported it into jpeg format. You see a huge jump in resolution; upwards of 118 for the MTF50 lp/mm measurement. Fake! Fraud! Bogus!
Jpg sharpened in LightRoom Amount = 50, Radius = 1.3
I jacked up the sharpening in this version of the same raw original, exporting it with the “Amount” parameter changed from 36 to 50, and increasing the “Radius” parameter from 1.0 to 1.3. The MTF50 now passes 120! Anybody who’s paying attention would start to get pretty suspicious about these measurements. Faker! Frauder! Boguser!
The Chart Up Close
Let’s take a look up close to see what’s happening in each shot.
Unsharpened Raw shot. Numbers are edge “cycles per pixel” measurement.
The close-up above is near the chart center, showing the edge MTF50 measurements in units of “cycles per pixel”. The measurement software overlays the measurements onto each of the edges. This is the raw-format shot without any image processing to adjust it. The measurement of 0.26 above, for instance, is an MTF50 of 60.1 lp/mm on this Nikon D850 sensor. In other units, this measurement is 2873 lines per picture height. If this shot was a landscape, the urge to sharpen it up would be overwhelming, but don’t!
The jpeg shot above is the same photograph, at the same chart location, but with minimal sharpening applied. The resolution measurements are hugely different, because the edges have a much shorter transition zone between black and white. The fuzzy grey zone between black and white is mostly removed. This is what makes sharpened photos look so much better than untouched raw versions. The measurement of the same edge has jumped from 0.26 c/p to 0.51 c/p, or from 60.1 to an astonishing MTF50 of 117.8 lp/mm. This same measurement is the equivalent of 5630 lines per picture height. Almost like getting your hands on some sort of advanced alien technology.
More aggressive sharpening makes the edge transitions even more abrupt, which translates into astronomically high resolution measurements. But those measurements are of no use to evaluate actual lens performance. The same edge here jumps to an MTF50 of 122.4 lp/mm, or 5851 lines per picture height! Outrageous!
There’s lots of bogus information out there on the internet. This is just another example of how that can happen, couched in the cloak of “science”. Editing tools like the “unsharp mask” definitely have their place in photography, but not when trying to analyze how sharp a lens is. As the old saying goes, “buyer beware”.