MTF with the naked eye

0
0
0
s2smodern

 DSC3618

A difference of 10% in resolution is simple to measure. Can you see that as well? Is a 10% difference something for you to worry about, or is that 20%? Try it yourself on the basis of the practice shots in this article: if you mouseover the picture, you can see the measured resolution.
You might wonder whether a measured sharpness corresponds with what your own eyes see in a practice shot. Manufacturers think so. In the design of a lens, manufacturers calculate the resolution (MTF) of lenses. The publish calculated – and a few publish measured – MTF values of a lens. They determine that MTF in different ways, so that it is difficult to compare the diagrams from different manufacturers with each other. In addition, lenses in practice are usually not quite as good as in theory.

Pictures can be enlarged or shrunken by your smartphone, tablet or computer. View the 100% image cut-outs from practice shots, preferably on a screen wider than 1200 pixels and with the browser display at 100%.

For every camera or lens test, we compare the practice shots with the measurements that we do. If we take the practice shots carefully, the conclusions from measurement and practice always correspond. We measure the resolution of lenses and cameras with the help of Imatest, so that you can compare the quality of different systems (lens, camera and image editing) with each other. That produces an enormous amount of measurement results. In order to keep it simple, we translate the lens quality into a score that varies from 6 to 9.9, where a higher score is better.
Photographers with a score fetish and/or eagle eyes worry about a resolution difference of 5% or less. CameraStuffReview is not trying to measure to as many places after the decimal as possible. We strive for our scores to correspond with what we and you can see with the naked eye. With a difference in resolution of about 10%, the scores of lenses are equal to each other.
Final scores in our lens tests are not averages of sharpness, vignetting, distortion and chromatic aberration. There are other factors that determine the lens quality, and we count those in the final score as well. The bokeh (background blur) of shots taken with a long focal length (read: a big sensor) is, for example, usually nicer than that in shots that are made with a shorter focal length (read: a smaller sensor). We work that difference in image quality into the final score for a lens.

In this article, we are going to discuss as little as possible the many ways that you can measure sharpness. There is of course a difference when you assess the sharpness by looking at jpg files (Photoreview), unedited RAW files (Lenstip, DxO), or sharpened RAW files (Photozone). For unsharpened RAW files, the measured resolution is lower, and if you sharpen carefully, it is higher than in jpg files. We limit ourselves to a couple of jpg files directly out of the camera. The trend is the same for RAW files as it is for jpg files.

Center sharpness and corner sharpness

2000vs500zondertekst

At full aperture, a lens can sometimes show a significant difference in sharpness between the center and the corners. Sometimes it is even as great as in the example above. Very bright lenses (over f/1.8) from a few years ago are clearly less sharp in the center at f/1.2 or f/1.4 than at f/5.6. A difference in resolution of a factor of 4 is, as you can see, easy to recognize. The resolution of the picture on the left corresponds with the center sharpness that we measured from the Canon EF 16-35 mm f/4L IS USM on a Canon 5D MK3 (full-frame sensor with anti-aliasing filter), Canon EF-S 10-18 mm f/4.5-5.6 IS STM on a Canon 650D (APS-C sensor), Nikon 1 18.5 mm f/1.8 on a Nikon J2 (1-inch sensor) or an Olympus 14-42 mm f/3.5-5.6 EZ ED M.Zuiko Digital on an Olympus OM-D E-M1 (micro-43 sensor without anti-aliasing filter). The resolution in the picture on the right is the corner sharpness that we find when testing less expensive lenses on a camera with less than 16 megapixels. Fortunately, we see such extreme differences less and less often. That is why we are showing a few practice examples of smaller differences in resolution.

Extremely high sharpness

2000vs4000zondertekst

What looked sharp in the previous picture now looks blurry. Examples of lenses that realize the highest possible sharpness are primarily lenses on a camera with a full-frame sensor like, for example, the Tokina 24-70 mm f/2.8 AT-X PRO FX SD, Canon EF 100 mm f/2.8L Macro IS USM (both tested on a Canon 5DsR), Sony FE 24-70 mm f/2.8 G Master, Zeiss Batis 25 mm f/2 on a Sony A7R II, or a Nikon 70-200 mm f/2.8G ED-IF AF-S VR on a Nikon D810. And of course various Sigma Art lenses like Sigma 20 mm f/1.4 DG HSM Art, Sigma 24-35 mm f/2 DG HSM Art or Sigma 50 mm f/1.4 DG HSM Art. Even so, there are also a number of lenses that are able to achieve this high resolution on an APS-C sensor, like, for example, a Nikon DX 16-80 mm f/2.8–4E ED VR on a Nikon D7200 or Samsung 16-50 mm f/2-2.8 S ED OIS NX on a Samsung NX1. You really need an APS-C sensor without anti-aliasing filter and 24 megapixels or more in order to approach or match the resolution of the best full-frame lenses.

The example above illustrates beautifully what happens when you put an old lens on a modern camera with 36 megapixels or more. The center of the image benefits from the higher resolution of the sensor, but in the corners, the sharpness does not increase at an even pace. It therefore suddenly seems as though your trusty old lens is no longer as good on a camera with high resolution. With good modern lenses, the sharpness in the corners is higher, and you do not find these kinds of differences. In the past few years, there has been enormous progress made in the design of bright lenses, including the Sigma Art lenses, Sony G Master and the Nikon f/1.8 series of fixed focal lengths, but also the professional lenses from Olympus and Panasonic stand out in particular to me.  

Technical intermission: lp/mm or LW/PH?

The pictures above are a great illustration of the extremes that we encounter in measuring resolution. These differences can not only be measured well, they are also recognizable with the naked eye. There are multiple ways to express the resolution. The most commonly used units are lines per picture height (LW/PH) and line pairs per mm (lp/mm). It is simple to convert back and forth:

  • A full-frame sensor is 24 mm high: 50 lp/mm = (2*24*50=) 2400 LW/PH
  • An APS-C sensor is 15 mm (Canon) of 16 mm (Nikon, Sony) high: 80 lp/mm = 2400 LW/PH
  • A micro-43 sensor is 13 mm high: 92 lp/mm= 2400 LW/PH
  • A 1-inch sensor is 9 mm high: 2400 LW/PH = 125 lp/mm

When you express the resolution in lines per picture height, then the resolution measured on cameras with the same image ratio can be compared directly with each other. That is why we use this unit. Because a micro-43 sensor with an image ratio of 4:3 is narrower and taller than most other sensors, with an image ratio of 2:3, we calculate the resolution scores for micro-43 lenses with an image ratio of 2:3. We thus make the test results for sharpness comparable for all lenses, regardless of the sensor dimensions.

The resolution in lp/mm of lenses on a smaller sensor is higher than that of lenses on a larger sensor.

If you want to know the quality of lenses, then you can eliminate the influence of the sensor size on the resolution by expressing the resolution line pairs per millimeter, instead of lines per picture height. The more line pairs per millimeter a lens can distinguish from each other, the better the lens. Lenstip expresses the resolution (of unsharpened RAW files) in lp/mm. On a camera with a 20-megapixel full-frame sensor (Canon EOS 5D Mark III), the best lens in the tests by Lenstip achieve a resolution of 47 lp/mm, and a good lens starts at about 30 lp/mm.
For 12-megapixel RAW files made with an Olympus E-PL1 (micro-43), the record stands at 83 lp/mm. Micro-43 lenses are only marked on Lenstip as good if they have a resolution of 45 lp/mm or better.

20% difference in resolution: Do you see the difference?

3300vs2600minizondertekst

Now and then, we are asked by readers whether we can give general guidelines for how great a percentage difference in resolution is relevant for them. I often give the answer of 20%. Honest said, that depends on many factors that we have no control over. At what enlargement are you going to view or use the image? Do you photograph by hand or from a tripod? If you photograph by hand, then the chance is greater that the 20% is applicable for you, since then you apparently are not as worried at that last little bit extra. With some designs, you see the difference in resolution more easily. See the test shots on DpReview sometime, where they compare the quality of cameras with each other on the basis of test shots from the studio. In some places, small differences can be seen clearly; in other places, it seems to make little difference when you compare a shot taken with a high resolution camera with a shot taken with a lower resolution. One of the causes for that can be that the high resolution shots by DpReview are reduced for print, so that you see the subject at the same size regardless of the resolution of the camera:

Sharp on screen or in print

If you want to compare the sharpness of two pictures with each other, it is tricky when the shots are made with cameras with different resolutions. When you enlarge the shots on your screen to 100%, then one shot is much larger than the other. You could apply three different methods:

  • View images at 100% on your screen and accept that you are comparing apples and oranges: comparing a large picture (high resolution) with a smaller picture (low resolution).
  • Shrink all shots to 8 megapixels: you thus simulate the image quality of a print at A4 size.
  • Enlarge all shots through interpolation up to the highest possible resolution and then sharpen a bit as a correction for the software interpolation.

For our tests, we apply the first and third methods, because we assume that printing is a separate area, in which much more is revealed than just shrinking to 8 megapixels. The quality of a print is also strongly related to image editing, sharpening, paper choice and much more. We try in our lens and camera tests to assess the quality of the original, without bothering with what you are going to do with it later.

3900vs2000minizondertekstYou have now seen a few practice shots with their associated resolution. Have you been able to determine the resolution of the pictures above with the naked eye? How accurately?

20% difference is sometimes worse than a 50% difference

2000vs1600minizondertekstWhat appeals to me about the assessment criteria on Lenstip is that a limit is set for adequate image quality and good image quality. That connects with my own experience: a difference of 10% or 20% is more important for me at low resolutions that at extremely high resolutions, where I really have to search carefully to see differences. The practice shots above are both made at a relatively low resolution. A 20% difference in resolution makes a great deal of difference there. But in the example below, the resolution is already a bit higher, and a 20% difference in resolution is much less of something to worry much about. 

Contrast-rich light or not/Contrast and micro-contrast

2900vs2500minizondertekst

When measuring sharpness, you can make a distinction between resolution (micro-contrast) and contrast. Lens manufacturers calculate the contrast for 10 lp/mm and the resolution for 30 or 40 lp/mm. The human eye combines those two. If you take pictures with a very good lens of a subject with a low contrast, then we assess those shots as less sharp than shots taken with the same lens under high-contrast conditions. You see that difference in these two shots. The two pictures are made with the same camera settings, but in the left picture, the sun broke through for a moment. When you measure the resolution as the MTF50 (in which the sharpness and contrast are combined in 1 measurement value), then you get a measurement result that correlates with the sharpness that we experience: the shot on the left is sharper.

Canon 5Ds vs 5DsR: Resolution with or without an anti-aliasing filter

4100vs3900zondertekst1200

Nearly all cameras have a sensor where each individual pixel can only see one color: red, green or blue. For converting a RAW file, the pixels peek at the other two colors from their neighbors. That works surprisingly well, although it delivers a slightly less natural picture than sensors where every pixel sees all the colors, like the Foveon sensor of Sigma. Only with very fine, regular patterns (sheer curtains, the fabric of a wedding dress, patterns in buildings) at the limit of what the resolution of a sensor can handle does a color interference pattern (moiré) sometimes occur. In order to prevent this phenomenon, cameras had an anti-aliasing filter that spread the light out over a few pixels. The extra blur resulting from the anti-aliasing filter is repaired with software by sharpening a RAW file, even before you see it on the screen.

Today, the resolution of cameras is so high that with a print at A4 or A3 size, you no longer see the individual pixels. If a camera does nothing to combat moiré, then you sometimes see that disruptive color pattern with a camera of 20 megapixels or less without an anti-aliasing filter in the details of an A3 or A4 print. But with a 50-megapixel camera, the resolution is so high that no longer see any moiré in an A4 or A3 print, even if you can see it on your screen when enlarged to 100% or more. In addition, you can remove moiré from a shot afterwards, just in the part where it is visible, so that an unnecessary sharpening step in the rest of the image is avoided.
What is the effect of an anti-aliasing filter on the sharpness? It makes a difference of about 10% in resolution. If you use a good lens, then the practice example above shows what difference you can expect between a shot taken with a camera without an anti-aliasing filter (Canon 5DsR) and a camera with an anti-aliasing filter (Canon 5Ds).

Influence of AF on the sharpness

3900vs3300minizondertekst

You could think that you are looking at a photo made with a good and a very good lens on the same camera. But you see an example here where a camera has focused twice in a row on the same subject, with the same lens and camera settings. The larger the sensor, the longer the focal length and the larger the aperture, the sooner you notice the difference in sharpness.
When testing lenses on SLR cameras, I have regularly encountered test shots in practice that were 50% less sharp than the other test shots made with the same camera and lens with the same camera settings. Anyone who reads the tests from Chip Foto video or Colorfoto carefully already knows that of course. This has nothing to do with front or back focus. With front or back focus, the camera always focuses too far forward or too far back. It has to do with the accuracy with which the camera can determine the right sharpness and the mechanical accuracy with which the focusing motor in the lens works. If you focus on the sensor signal, as in the case of an SLR camera with Liveview, then the spread in the sharpness as a result of AF errors is smaller. The mechanical quality of a lens and the length of the focus arc also play an important role. For accurate manual focusing, you need a lens with a very long focus arc (a turn of a focus ring of less than 1 mm makes a big difference for anyone striving for the highest sharpness).

Ivo Freriks
Author: Ivo Freriks
With Camera Review Stuff I hope to make a modest contribution to the pleasure that you get from photography. By testing cameras and lenses in the same way, evluating the results and weighing up the pros and cons, I hope to help you find the right camera or lens.

0
0
0
s2smodern

Comments (1)

  1. Robert

Congratulations! Excellent article.

 

Leave your comments

Posting comment as a guest.
Attachments (0 / 3)
Share Your Location