50 megapixels is not enough
To double the resolution,
you need 4 times as many pixels
At the start of the megapixel race, you could see clear differences in resolution when a new camera came out. A shot made with a 2-megapixel camera is visibly less sharp than a shot made with an 8-megapixel camera. The resolution of an 8-megapixel camera is twice as high as that of a 2-megapixel camera. That is a difference you can clearly see with the naked eye. For the same reason, the resolution of 4K video (with 8-megapixel frames) is an enormous advance relative to Full HD (2 mp).
The higher the number of megapixels, the smaller the influence of an extra 10 megapixels on the resolution:
- In order to double the resolution relative to a 6-megapixel camera, you need a 24-megapixel sensor.
- In order to double the resolution of a 24-megapixel camera, you need a 96-megapixel sensor.
If you compare the sharpness of a 16-megapixel camera with the sharpness of a 20-megapixel camera, then the influence of the number of pixels on the sensor is less than the effect, for example, of the lens quality, or perhaps the (lack of an) AA filter in front of the sensor.
Since the arrival of the Canon 5Ds with its 50-megapixel sensor and the Olympus OM-D E-M5 mk2 with 40-megapixel, high-resolution shots, it seems as though the megapixel race has begun again. Do we see a difference between a shot made with a 36-megapixel camera or a 50-megapixel camera?
A few lenses good enough for a 50-megapixel sensor
At f/0.95, a good lens loses 1% in sharpness due to diffraction and perhaps 50% due to lens errors
Are modern lenses able to make the high resolutions of these new cameras visible? For an extremely good lens, the 50-megapixel full-frame sensor of the Canon 5Ds or Canon 5Ds R is no problem. More about that later. Perhaps even more interesting questions are: What do I have to do to realize the resolution difference between a 36-megapixel or a 50-megapixel camera in practice? Or: Is that difference worth the effort and the higher costs for me? And what do you think about: Do I see a difference in resolution in my photos if I use my current lenses?
There is a whole arsenal of lenses that actually produce shots with the Olympus OM-D E-M5 mk2 in the high-resolution mode with astronomically good resolution. But what is that like with a 50-megapixel, full-frame sensor? If you choose a camera with a 50-megapixel, full-frame sensor, you also have to replace a number of your lenses. Working from a tripod and focusing accurately become more important as the resolution of the camera becomes higher. Not every photographer wants to do that. So not everyone needs a 50-megapixel camera, either.
Whether you see the difference in sharpness between two pictures depends on:
- quality of camera and lens (megapixels, AA filter or not, glass types used, AF accuracy, etc.)
- camera and lens settings when taking the picture (aperture, shutter time, ISO, tripod, image stabilization, self-release, etc.)
- weather conditions and lighting (bright or dim light, for telephoto lenses: air turbulence)
- image editing (sharpening, contrast, noise suppression, de-mosaic of the RAW converter, correction for vignetting or distortion, etc..)
- quality of the print or the settings of the monitor
- size of the image and the viewing distance
- quality of the eyes and/or glasses of the viewer
What is sharp?
The larger the picture and the smaller the viewing distance, the more easily you see a difference in resolution. Billboards are printed in low resolution, but because you view them from a big distance, they still look sharp. A poster with a resolution of 150 pixels per inch (ppi) might look just as sharp as a glossy magazine printed in a resolution of 300 ppi. You view the poster from a much larger distance than the magazine. What is sharp?
Below you see two image excerpts from test shots with about 10% difference in resolution, shown at 50% of the actual size. With the naked eye, it is difficult to see differences between the two shots, despite the fact that these two image excerpts are significantly enlarged (the exact dimensions depend on the resolution of your screen). And you are viewing them from close-up!
Sharpness loss due to diffraction at f/2.8 : 7% (36 mp full frame) to 12% (1 inch 20 mp)
If you are not checking pixels and you seldom make large prints, then it probably does not make much sense for you to choose a camera with more than 24 megapixels. How great the difference in sharpness is that you see with the naked eye differs per person. With direct comparison of two pictures next to each other, you can assume that most people will see a difference of 20% in resolution. If you view photos on a screen at 100%, then you can see that a Canon 5Ds has higher resolution than a Canon 6D or 5D mk3. The comparison of a Canon 5Ds with a Nikon D810—even at 100%—is much more difficult.
Try it yourself: If you sharpen in Photoshop with Unsharp Mask in two steps (first radius 0.8 pixels, 85%, and then a radius of 20 pixels, 10%), then the shot looks visibly sharper and more natural than when you only sharpen the micro-contrast (radius 1) or the global contrast (radius 20).
Our eyes experience high contrast—a large difference in brightness between the darkest color and the lightest color of something we are looking at—as sharp. Sharpening software does nothing more than increase the contrast of a shot in a very clever way. Photo editing programs often distinguish when sharpening between the global contrast of a photo (the difference between the lightest and darkest point in the whole photo) and the detailing (local contrast differences, also called micro-contrast).
Seeing sharpness: contrast and micro-contrast
The finer the details, the lower the contrast.
Our eyes experience an image as sharp if the contrast is high and the details are clearly visible (“micro-contrast”). A silhouette of someone on a sand dune (high contrast) is sharp, but in this case contains little detail. A fine pattern of branches or leaves made with the same camera and lens at the same focal length and the same aperture as the shot of the silhouette on the sand dune has a high resolution, but looks less contrast-rich than the silhouette. If the pattern becomes even finer, like individual grains of sand in the sand dune, then we no longer see the contrast difference. If you view details at the limit of what a lens or camera can handle, then you will be squinting in dark glasses with extremely high resolution and extremely low contrast.
Resolution measurements: MTF10 and MTF 50
At f/4, the loss of sharpness due to diffraction is 20 to 35 %. Sometimes, lens errors have an even effect than diffraction.
There are different ways to measure resolution. Colorfoto measures, for example, the resolution at which the distinction between black and white is practically invisible (MTF10: the difference between the darkest and the lightest color is just 10%). This will also be identified as “vanishing resolution,” and this measurement value comes the closest to the number of pixels on the sensor (sometimes identified as the Nyquist frequency). Because our eyes experience the combination of contrast and details as sharp, only determining the vanishing resolution is not sufficient if you want to know whether a lens is sharp or not. Colorfoto therefore also measures the global contrast.
Most lens testers (including CameraStuffReview) use the MTF50 (MTF50: the difference between the darkest and the lightest color is 50%), because it is assumed that this contrast measurement correlates best with the sharpness that the human eye experiences. That corresponds with our experiences: a photo made with a lens with a high MTF50 looks sharper than a photo made with a low MTF50. If all other conditions—test camera, distance to the subject, shutter time, aperture, ISO, etc.—are otherwise the same, of course.
For the highest center sharpness on cameras with a high pixel density, go no higher than f/4 – f/5.6. At f/5.6 a micro-43 lens (just like a lens on a Canon 5Ds) gives up 30% to diffraction.
What resolution does a lens have to be able to handle in order to let the resolution of a Canon 5Ds R come fully into its own? At least a vanishing resolution that is equal to the number of line pairs per mm of the sensor (0.5*5792 pixels/24 mm = 119 lp/mm). More is better, of course. But how do we know that a lens has such resolution, when this is the first 50-megapixel, full-frame sensor?
If we limit ourselves to center sharpness, then we can find the answer in the near future with lenses tested on the Canon 750D/760D (with AA filter: 134 lp/mm). Because the Canon 5Ds R corresponds as far as pixel density is concerned with a Nikon D7200, and neither camera has an AA filter, it is not crazy to assume that the best lenses on a Canon 5Ds R can come out at a resolution of 115 lp/mm or more. Hence the title: 50 megapixels is not enough.
If a lens performs well on a Nikon D5500, D7100 or D7200, then a high center sharpness with the Canon 5Ds (with AA filter) or 5DR (without AA filter) is a breeze. The Sigma 35 mm f/1.4 Art is a good candidate. This is one of the lenses with—tested on a camera with a 130 lp/mm sensor—the highest center sharpness that we have seen. The Tokina 11-16 mm PRO DX II, the Sigma 18-35 mm f/1.8 Art and—to my surprise—the Nikon 18-140 mm all score very high in our tests with their center sharpness, but they are not designed for a full-frame sensor.
Diffraction and resolution (in lp/mm)
At f/8, there is still 60% left of the maximum possible sharpness. A 1-inch sensor at f/8 delivers only 40% of the highest possible sharpness.
Diffraction is a natural phenomenon: light becomes more scattered as the opening it passes through becomes smaller. In terms of diffraction, you make the get the sharpest picture at f/0.95 and the least sharp picture at f/22. With a pinhole camera, the aperture is so small, not seldom f/100, that the light is spread out over multiple pixels. You therefore never get a sharp picture with a pinhole camera.
If you take the size of the pixels and the influence of diffraction into account but do not consider lens errors, then the resolution of a camera is simple to calculate for every aperture. You see the results in the illustration above.
At f/0.95, a lens without lens errors and a camera without an AA filter would produce the same vanishing resolution (MTF10) as the number of rows of pixels on the sensor. A Canon 760D (sensor height: 14.9mm; 134 lp/mm) or a Nikon D7200 (sensor height:15.4mm; 130 lp/mm) both have a sensor with a height of 4000 pixels. At f/0.95, the influence of diffraction on the resolution is negligible, and a lens without lens errors would be able to deliver a resolution of 132 lp/mm (Canon) or 128 lp/mm (Nikon). That is higher than the 119 lp/mm of the Canon 5Ds or the 101 lp/mm of the Nikon D810. The very best lenses on a camera with a full-frame sensor do not achieve, expressed in lp/mm, such high values. On a Nikon D800, with a 36-megapixel sensor, the Sigma 105 mm f/2.8 macro reaches a vanishing resolution of 92 lp/mm (2200 line pairs per picture height). On the Nikon D7100, the Tokina 70-200mm F4, in a recent test by Colorfoto at a focal distance of 70 mm and f/8, achieved a resolution of 1800 line pairs per picture height, which corresponds with 117 lp/mm. In both cases, the sensor can still be a limiting factor. They probably score higher if you test them with a 50-megapixel, full-frame sensor.
Micro-43 scores higher than full-frame (in lp/mm)
For Lenstip and CameraStuffReview, the MTF50 of unsharpened RAW files is measured; that is lower than the vanishing resolution (MTF10) that Colorfoto reports. CameraStuffReview uses lines per picture height (I will come back to that.), and Lenstip reports measurement results in lp/mm. Lenstip makes heavier demands of the cameras with a smaller sensor:
- For a 20-megapixel, full-frame camera (Canon EOS 5D Mark III), the resolution of a lens according to Lenstip is sufficient at 30-32 lp/mm..
The best lenses on this camera achieve an MTF50 resolution of 44-47 lp/mm.
- For a 12-megapixel micro-43 camera, Lenstip utilizes much stricter criteria (in lp/mm) than for a full-frame sensor: At 42 lp/mm—considerably more than the norm for a full-frame lens—on a 12-megapixel micro-43 camera that is sufficient according to Lenstip.
The best micro-43 lenses achieve a resolution of 80 lp/mm (with a record of 82.6 lp/mm for the Voigtlander 0.95/25).
Even if you take into account the unusual 3:4 format in comparison with the 2:3 format of a full-frame sensor, it seems unfair to set higher requirements for a smaller sensor than for a large sensor. Calimero would protest: "They are big and I is small and that is not fair, oh no!".
Bright Nikon 1 lenses are the record-holder with 150 lp/mm
A perfect f/0.95 lens should in theory be able to achieve a maximum resolution of 200 lp/mm on a Nikon J5 at f/0.95. That is twice as high as the highest possible resolution of a full-frame camera. If you want to know who the resolution king is, if you express the resolution in lp/mm, it would be one of those. Nikon demonstrates with the Nikon 1 18.5 mm f/1.8 that a good lens in practice can achieve a vanishing resolution (MTF10) of 150 lp/mm: Colorfoto is one of the few labs that determines the resolution of a lens at the point (the vanishing resolution) where the resolution is so high that the contrast has practically disappeared. Colorfoto reports for the Nikon 1 18.5 mm on a Nikon V2 a vanishing resolution of 1328 line pairs per picture height. The Nikon 1 sensor has a height of 8.8 mm, so this corresponds with a resolution of 150 lp/mm. The Nikon 1 18.5 mm f/1.8 scored very high in our tests as well. Just like the Nikon 1 32 mm f/1.2, which is perhaps even better.
Resolution in lines per picture height (LW/PH)
Forget about lp/mm. LW/PH corresponds better to practice
If you express measured resolutions in lp/mm, then you cannot simply compare measurement results with each other. If you want to compare the measured resolution of lenses on different cameras with each other, then you have to start with a resolution that is independent of the sensor height and image ratio. But digital cameras have sensors with different sensor heights and image ratios:
|sensor||image ratio||sensor height||camera|
|Full-frame||2:3||24 mm||Nikon, Canon, Sony|
|DX||2:3||15.4 mm||Nikon, Samsung, Sony|
|Micro-43||2:3||13 mm||Panasonic, Olympus cameras|
|1 inch / CX||2:3||8.8 mm||Nikon 1, Samsung mini cameras|
If you express the resolution in lines per picture height, then it is as though you are comparing photos with each other that are all printed at the same size or shown at the same size on a screen. At CameraStuffReview, we make the image ratios the same by using a micro-43 camera in the 2:3 image ratio and expressing all resolutions in lines per picture height (LW/PH). If you take diffraction (but not lens errors) into account and consider the maximum achievable resolution of different cameras expressed in lines per picture height, then you see that the Canon 5Ds is the ideal camera right now for realizing the highest possible resolution.
f/11 gives great focal depth, but diffraction costs you 50% of the sharpness
In theory, an ideal lens without lens errors at f/0.95 on a sensor without an AA filter would have the same vanishing resolution as the height of the sensor (5792 pixels). But an ideal lens does not exist. In practice, most lenses are not entirely sharp at full aperture. By stopping down a few stops, you eliminate the lens errors and achieve the highest possible sharpness. The sharpness subsequently decreases as a result of diffraction. How relevant are the resolution differences that you see between the different cameras in the picture above?
Judge for yourself: You see a 20% difference in resolution with the naked eye
f/16 and f/22 are better avoided if you want center sharpness. Unless you need a great deal of focal depth.
If you compare two shots with a difference in resolution of 20% by displaying them at 100% on a screen next to each other, then you see the difference in resolution on the monitor of a desktop computer with the naked eye. On a smartphone, the picture is too small to be able to see the difference. Move your mouse over the picture below and judge it with your own eyes.
At f/2.8, a lens on a Canon 5Ds (50 mp) delivers a maximum of 50% more resolution than on a Canon 5D mk3 (20 mp). At f/22, that difference is only 5%.
Hopefully, based on the explanation above, it is clear why bright lenses are so often renowned for the high sharpness that they offer at f/4 or f/5.6. If you stop down two stops with an f/1.4 lens, then you have significantly reduced most of the lens errors, while the negative influence of diffraction is still relatively minor. That is why the most modern bright lenses, like the Sigma 18-35 mm f/1.8 Art (for cameras with a DX or APS-C sensor) or the Sigma 35 mm f/1.4 Art (for full-frame/FX) are ideal candidates for demonstrating the high sharpness of 24-megapixel APS-C cameras or 50-megapixel full-frame cameras. In the center, since very few lenses manage to achieve the same sharpness in the corners as in the center at these kinds of extremely high resolutions.
Modern diffraction-limited lenses of f/4 and f/5.6
Only extremely good, but less bright (f/4, f/5.6) modern lenses are so well designed that they produce the highest sharpness even at full aperture. As far as the center sharpness is concerned, you do not have to choose a smaller aperture with these lenses. We saw that when testing the Sigma 24-105 mm f/4 Art DG OS HSM, Nikon AF-S DX 18-140 mm f/3.5-5.6 G ED VR, Canon 24-70 mm f/4 IS and the Tokina 70-200 mm F4. The center sharpness of these modern lenses is maximum at full aperture and only decreases as a result of diffraction as you stop down. For the corners, it is sometimes a bit more nuanced: the sharpness in the corners usually still increases a bit if you stop down 1 or 2 stops.