Is 24MP actually matched to Full-HD Resolution when factoring in the differences on how pixels are defined?

Naive assumption: 24MP sensor vs a Full HD monitor of 2MP, sensor has much more resolution.

Second look: You sample something that is not synchronous to your sampling raster/clock, you need to respect Nyquist - so you either lose something to antialiasing or to aliasing, you end up with only half the resolution in each dimension guaranteed. Now we are at 6MP vs 2MP (unless the display actually antialiases! Also, a monitor can synchronize the pixels to their clock, a sensor can't synchronize with reality).

Trying to be too clever: A sensor defines pixels as monochrome, one R or G or B value, while the monitor is defined in triplets of monochrome pixels. Now we are at 2MP vs 2MP.

Is that the underlying reason why 24MP has become a quasi standard?

One underlying question is: If you shoot a lens that would be good for 2MP on a 24MP sensor and display it on a "2MP" screen, the result will definitely look flawed. Why?

• 24MP is 6000x4000. "Full HD" is 1920x1024, barely 2MP. I would doubt many serious photographers use as low a screen res as "Full HD" for editing. Not a clue what you're actually asking, sorry. Commented Sep 17, 2019 at 19:29
• Your math is flawed. Bayer filter has 4 colored pixels per group: one red, two green, one blue. So, if there was any truth in what you claim, the quasi standard would have to be 32 MP. Commented Sep 17, 2019 at 20:00
• Also, I'm not entirely convinced that an anti-aliasing filter would halve resolution in each dimension. Do you have any sources to back that up? Nyquist means the maximum frequency is half of the sampling frequency, but it doesn't really apply here... Commented Sep 17, 2019 at 20:02
• Half frequency should apply to both temporal and spatial sampling, no? Commented Sep 17, 2019 at 20:16
• @Tetsujin you'd assume it is still what most photos are displayed at today, wherever they are edited :) Also, one loses all hope when looking at how some graphics-related professionals don't find anything wrong with high-res but uncalibrated, in the worst case viewing angle dependent, displays.... Commented Sep 17, 2019 at 20:19

No, the truth is somewhere in between. You're partially right, but you're missing a few things.

Regarding spatial sampling, bringing up Nyquist is mostly a red herring, because you can make the same exact argument about the discrete pixels of monitors as you can about sensors. So one pixel of sensor is as good as one pixel of monitor, for this purpose. However, when cameras use optical low-pass filters (anti-aliasing filters) to avoid aliasing effects on the sensor, this does reduce the spatial resolution of the sensor slightly (due to the impossibility of a perfect brickwall filter), slightly reducing the information content of each pixel (because pixels are correlated with their neighbors, therefore knowing the neighbor provides partial information about the pixel).

Regarding the Bayer filter... yes, it's absolutely true that each color is sampled at a lower resolution (in the classic RGBG filter, 1/2 linear resolution for red and blue, and 1/sqrt(2) linear resolution for green). And this necessarily decreases the maximum information captured by a Bayer-filtered sensor of a given pixel dimension, compared to what could be displayed by a full-RGB monitor of the same pixel dimension.

But it's important to remember that the scenes we photograph aren't maximum-entropy sources. To whatever extent the scene contains mostly low-frequency information, below the Nyquist rate, the AA filter removes very little information — and mostly they do. To whatever extent the color channels in a given patch of the scene are correlated in a predictable way, the Bayer filter removes very little information — and mostly they are. So the information-theoretic "effective resolution" for an image sensor photographing ordinary scenes is much higher than what you calculate it to be. Without any real mathematical backup, I'd say that it's probably 75-90% of the raw pixel count, not 8%.

Given this, and the lack of any supporting evidence (and the aspect ratio difference pointed out in another answer), there's no reason at all to think that the prevalence of 24MP sensors has any relation at all to 1920x1080 monitors or 1080p video formats. More likely, it's just a convenient place where manufacturing technology, consumer expectations for decent photos, and price all come together.

• OLPF-less cameras tend to use software antialiasing AFAIK.. Commented Sep 18, 2019 at 8:11
• +1 for underlining that 1 pixel of monitor = 1 pixel of sensor, regardless of Nyquist. Commented Sep 19, 2019 at 8:45

Explanation

Aspect ratio

6000x4000 pixels is a 3:2 aspect ratio.

1920x1080 pixels is a 16:9 aspect ratio.

So the number of pixels from a typical 24 MP sensor used to produce an HD image are not all 24M of them. Even if the width is "uncropped", only 3375 of the 4000 horizontal lines are used. That's a total of 20.25 MP.

"Raw" vs. RGB

Let us say "Red" ("R"), "Green" ("G"), and "Blue" ("B") are the colors on which the filters in a Bayer mask are centered and most transmissive.

Let us label the frequencies of light used in RGB color reproduction systems as R, G, and B.

Raw image data is monochrome, but each "pixel" does not exclusively collect "red", "green", or "blue" light. Each filter lets through a wide range of frequencies with a peak at a particular "color". There is a lot of overlap of what gets through each color filter, particularly between the "green" and "red" filters. All three R, G, and B values are interpolated. "Red" filters in a typical Bayer mask are centered at about 590 nm, which is more yellow-orange than red. The Red channel in RGB is at about 640nm. The Bayer mask filters are not centered on the G and B frequencies used in RGB, either.

The R, G, and B values for each pixel must all be interpolated from the raw values of the sensels covered with "R", "G", and "B" filters because "R" ≠ R, "G" ≠ G, and "B" ≠ B.

Each of the filters in a Bayer mask allow a wide range of wavelengths through. They are attenuated for a peak transmission at about 455nm ("Blue"), 540nm (Green), and 590-600nm ("Red"). There is also a lot of overlap between what gets through each filter compared to the others. Our RGB color reproduction systems use values of around 480nm (Blue), 525nm (Green), and 640nm (Red) for the three primary colors. Some screens also include Yellow subpixels emitting at about 580nm. As you can see, the peaks of the detectors used in our cameras do not match the colors used in our output devices.

For more, please see:Why are Red, Green, and Blue the primary colors of light?

Conclusion

The result of demosaicing/interpolation of a 24 MP raw file is normally a 24MP image with 24,000,000 RGB pixels, not an image with 12,000,000 G, 6,000,000 R, and 6,000,000 B pixels. It's also usually in a 3:2 aspect ratio (6000 x 4000 pixels), not a 16:9 aspect ratio.

• What confuses me: A bayered sensor would be unable to acquire some pathological color patterns that a display of the same resolution could easily replay. I am aware that humans don't see color full res anyway, but I didn't START marketing with apparently useless specs :) Commented Sep 17, 2019 at 23:12
• Bayer sensors imitate the human retina fairly closely in a lot of ways. What the marketing department does with that is a whole other can of worms. Commented Sep 18, 2019 at 0:29
• +1. I don't understand why somebody downvoted this. It's pefectly valid info, and well related to the question. Commented Sep 19, 2019 at 8:47