[UncleVanya]

A great question. If you understand what diffraction is, because you took high school physics, the idea that you can compensate for it is somewhat amusing. I suspect it's some kind of slight of hand effect that accelerates micro-contrast and sharpness, that is labelled "diffraction compensation" that does what my software does when I ask it to. When I look at my true duration limited images, most taken with very small sensor cameras.... there's just no fixing it. The problem being if there was a real way to reduce direction, it would have to involve some kind of polarizer that eliminated the light diffracted off the edges of the aperture blades, but didn't interfere with directly transmitted light. It would be mechanical, not a software fix.

Buried in this thread ( https://m.dpreview.com/articles/1884561533/pentax-firmware-v1-10-for-k-3-add...ion-correction ) there’s some technical discussion. The claim is that the info isn’t lost until about f11 and that with a clean low iso signal well understood predictions can restore some of the information. I’m continuing to look into it.

[Papa_Joe]

I would guess it is for manufacturing reasons. More delicate chips are prone to get a higher rate of defective ones. Probably 24 MP are now the sweet spot in the profit margin.

[normhead]

Buried in this thread ( Pentax firmware v1.10 for K-3 adds diffraction correction: Digital Photography Review ) there’s some technical discussion. The claim is that the info isn’t lost until about f11 and that with a clean low iso signal well understood predictions can restore some of the information. I’m continuing to look into it.

I always say, don't tell me, show me, in this kind of scenario.

[UncleVanya]

I always say, don't tell me, show me, in this kind of scenario.

They mentioned threads that showed images and described software used in astrophotography that uses the same principles. I’m going to keep looking.

[ChristianRock]

I find 16MP gives enough detail for large prints. Even 10MP seems enough for me because you shouldn't really "pixel peep" prints, you should look at them as a whole image.
24MP files are still manageable even with older computers, they have more than enough detail and will allow quite a bit of cropping.
Also, I bet that 24MP full frame cameras still outsell the ones with more megapixels.

As an aside, I really like the 20MP sensor in my K-S1. Plenty of detail, lovely colors. I would say 16-20MP is plenty unless you really have a unique use case. I can see why the DSLR monster pro cameras (1Dx III, D6) are 20MP. As clackers said it, big beautiful pixels...

[aldo taddia]

I assume it's the right number of pixels to get a good photo result. It would not explain why you continue to make FF machines with 20 megapixels

[UncleVanya]

I assume it's the right number of pixels to get a good photo result. It would not explain why you continue to make FF machines with 20 megapixels

The larger pixels can be more friendly to low light. The larger pixels also permit more stopped down shots without diffraction.

[aldo taddia]

I assume it's the right number of pixels to get a good photo result. It would not explain why you continue to make FF machines with 20 megapixels

[UncleVanya]

I assume it's the right number of pixels to get a good photo result. It would not explain why you continue to make FF machines with 20 megapixels

We are not communicating.

A ff 20mp has much larger pixels that do better in low light than an apsc 24mp camera in general. Sensor she matters as well, newer tech is able to squeeze more performance (to a degree) from smaller pixels.

The choice of sensor size and megapixels is reasonably rationale but the marketplace also factors in. 20-24mp seems to have established at both apsc and ff as a floor on the megapixels or non-specialized cameras.

[ehrwien]

From Imaging Resource supporting the previous explanations from members:

"However, like most Canon DSLRs, the 90D's sensor includes a fixed optical low-pass filter. And while that's handy for combatting ugly moiré and aliasing artifact, it robs some of that finer detail resolving power, and we just don't see as much resolution and detail from this 32.5MP sensor as we'd have hoped. Canon gave us more megapixels, but they are hampered by the low-pass filter.

As the ISO rises, we run into another issue with this higher-resolution sensor, albeit not a severe one by any means. Thirty-two megapixels is a lot of pixels to cram onto a relatively small APS-C sensor size. Generally, all things being equal, the smaller the pixel size, the higher the noise for a given ISO compared to a similarly-sized sensor with a lower resolution (and thus larger pixels). As such, we do see more noticeable noise in 90D images, even at relatively mid-range ISOs, like 800-1600. As mentioned, the 90D is certainly not bad at high ISOs, and we found images are perfectly acceptable even up to around ISO 6400-12,800. But, the Canon 90D is certainly not class-leading when it comes to high ISO performance."

Doesn't the dual pixel sensor technology effectively double those 32.5 MP, not for resulting resolution, but with respect to "lost" light in between pixels, for the sake of being able to use every pixel for on-sensor PDAF?

[Wasp]

Modern Pentax bodies have pixel shift. A sensor with more than 24M pixels could mean that shifting the sensor by one pixel becomes mechanically impossible. The sensor is also moved ever so slightly to act as a low pass filter and the same applies there.

The other thing is, more pixels cost more money. The K3 III seems to be doing well but $2000 is probably the upper limit of what the market is willing to pay for a crop factor body. Only Leica charges more, for the $3200 24MP CL (pixie dust included).

Don't get me started on the prices of the top end MFT offerings from Olympus. The market sent them a loud message about that and they changed owners.

Canon is the current top dog in crop frame, with 32MP on the EOS 90D that retails for $1600.

[5ks]

Yeah, the more pixels you try to cram into a sensor, the smaller each one is.

That means that in darkness or shadows, not many light photons are being captured. By turning up the amplification to brighten the image, the noise is revealed. You'd have to compensate, say, by doing software noise reduction, which kills details and texture. I'm sure you've experimented cranking up noise reduction in Lightroom or Photoshop and seen people become wax dummies.

The K-5 and K-1 have big, beautiful pixels!

I agree a lot with clacker's post. Ca. 20 Mpix for APS-C and 36-40 Mpix for FF might be just optimal in pixel density, to provide us with "beautiful pixels" even with rather high ISO values. But, on the other hand, I haven't been using K-3 mk3 yet...

[AstroDave]

The claim is that the info isn’t lost until about f11 and that with a clean low iso signal well understood predictions can restore some of the information. I’m continuing to look into it.

Well, you can sorta cheat.

Radio astronomers have been using the CLEAN technique (https://en.wikipedia.org/wiki/CLEAN_(algorithm)) for quite some time. The gist is to find the peak in an image, and subtract some fraction of what you know the imaging device point source response shape looks like (the diffraction spot, aka Airy Disk, for an optical camera) from the image, centered on that peak position. The typical radio astronomy interferometer beam map is quite messy, in terms of sidelobes, and this technique works AMAZINGLY well. It was an integral part in making that "first map of a black hole" for the galaxy M87 a while ago. (As an aside, I was the first person in the world to actually KNOW that M87 had a compact radio source in it, back around 1970 or 1971 when I was a graduate student manning the night shift at the computer center where we were processing our VLBI (Very-long-baseline interferometry - Wikipedia) data tapes.)

A similar scheme was used to improve the early images from Hubble Space Telescope, back when it first went into operation with its mis-figured mirror and produced some rather ugly images.

How well this all works depends on the signal-to-noise ratio, and how well you really know the point-spread function (PSF). i.e. for our cameras, the apertures are not round (which is what an Airy disk is calculated for), so a more complicated PSF is needed to dig very deep. And, it probably varies across the image, as a function of look angle back out through the iris and lens.

[UncleVanya]

How well this all works depends on the signal-to-noise ratio, and how well you really know the point-spread function (PSF). i.e. for our cameras, the apertures are not round (which is what an Airy disk is calculated for), so a more complicated PSF is needed to dig very deep. And, it probably varies across the image, as a function of look angle back out through the iris and lens.

Given that this only works for lenses Pentax has profiles for I assume they have good understanding of the actual PSF of the lens at a given aperture and how this interacts with the exposure settings. (Aperture)

[clackers]

Diffraction is the same regardless of pixel size.

Definitely not!

Diffraction is strongly influenced by pixel size.

The smaller the pixels, the more likely an Airy disk smears across them.