How sensor formats matter?

[Erik Kaffehr]

Intention here is to discuss how formats differ, possibly viewed from an engineering point of view.

Back in say 2006 sensors were expensive. Most somewhat affordable digital cameras used APS-C size sensors and full frame 24x36 was rare. Most vendors used CCD sensors, with Canon being the sole champion for CMOS.
There was a small market for very expensive MFD backs, costing like 20-30k$US using CCD sensors in sizes up to 37x49 mm. (Later 54x41 mm).

Things are now different, all modern digital cameras I know of use CMOS sensors, mostly made by Sony. A smaller medium format size is now quite affordable. There is even a set of cameras using essentially the same pixel design, in all sizes from APS-C to 54x41 mm.

We may ask, how does sensor size matter? Let's compare the sizes:


Sensor sizes compared.

The most interesting numbers are the area factor, that decides how much the sensor can collect and 'Relative magnification' that is a parameter for achievable sharpness.
Let's take a look at the magnification stuff.


This is MTF I measured near axis on two lenses of mine. Blue is my Sony 90/2.8G on my A7rII and red is my Sonnar 180/4 on my 37x49 mm P45+ back. Note that the Sony lens has higher resolution in absolute terms.

I don't have official MTF data for the Sony lens, but both Hasselblad and Zeiss used to have measured MTF data for their lenses.


Looking near axis and at f/8 the figures are around 0.91, 0.83 and 0.65. Very close to my measured data. Could be coincidence... Or, it may be that my data is pretty accurate.

We have seen that the Sony lens is sharper, in absolute terms. But let's look at the magnification thing. I would normally print at A2-size, that is around 16"x23".
It is often regarded that about 180PPI is needed for an excellent print. The figure corresponds to 20/20 vision at 50 cm viewing distance. So, the number of vertical pixels needed would be 16*180 -> 2880. For that the 24x36 mm format would need 60 LP/mm while my 37x49 mm sensor would need 39 lp/mm.


So, here we can see that the Sonnar on the P45+ would deliver 66% modulation while the 90/2.8G on the Sony would have around 58% modulation. That may be a visible difference. What we see is that the larger format needs less magnification.


We can use line pairs (or cycles) per 'picture height' instead, that takes the magnification into account. The 1443 cy/PH figure is very close to our 2880 pixels needed for 16"x23". Note that red and blue lines are overlapping on the left side. The left side is dominant for visual 'acutance' so viewed at 'normal' viewing distances these images may be perceived as equally sharp.


This shows a 'selection histogram' covering the grey fields of the ColorChecker. Each field shows up as a 'bell curve'. The bright patches have narrow 'bell curves' and the dark patches have wide 'bell curves'. Note that each bell curve covers hundreds of different data numbers.

The 'bell curves' are caused by what is called shot noise, the random sampling of photons. That means that more photons captured improves the signal to noise ratio.
If we have a larger sensor, it can detect more photons at the same exposure. This may not be very noticeable in pictures shot at base ISO in good light. But assuming that all other things are equal, we would be able to use higher ISO or underexposure more on a larger sensor.


Here we compare Fujifilm X-T4 pushed 4EV, 24x36 mm pushed 5EV and Fujifilm GFX 100 pushed 6EV. The results are similar. Unfortunately, the GFX 100 shows a banding pattern as the firmware tries to hide possible artifacts from the metal masks delivering PDAF functionality. We can see that the noise levels are similar.

Technically, rising the ISO is just underexposure, so we could use the GFX at higher ISO than say the A7rIV. Also, we may need to reduce exposure to protect highlights. A larger format would allow us to reduce exposure more and still keep good detail in the dark parts of the image.

Let's look at image quality:


This is simulated A0 at 180 PPI. The crop show detail good for judging sharpness and resolution. The Fujifilm X-T4 yields remarkably good results for 26MP on APS-C. It may be that the GFX 50 is a bit to sharp. That would probably be do the the reduced pixel aperture on the GFX 50 sensor. It would need less sharpening. Here I kept sharpening the, though.

Now let's look at aliasing:


In a way, Fujifilms X-T4 with it's X-trans stands out. It avoid color aliasing. The Sony A7III has a lot of color aliasing while the effect of the OLP filter is quite visible. That camera needs smaller pixel. The Sony A7rIV still shows false color. Fujifilm GFX 100 and Phase One 3100MP yields best detail. The Fujifilm GFX 50 shows heavy color aliasing, a combination of large pixels, sharp lens, small pixel aperture and no OLP filtering. But, this kind of subject is rather rare.

Reading the fine print...



The observations I may make is that:

• The Fujifilm X-Trans concept may make sense. The APS-C performs remarkably well.
• Having more megapixels may be helpful when printing large.

Best regards

Erik

--
Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic uses to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

 

[NAwlins Contrarian]

Erik, this is a minor point, but:

APS-C: the common "APS-C" film sensors are 23.5x15.6mm, and the Canon "APS-C" sensors are 22.3x14.9mm. The 25.1x16.7mm dimension I think only applies to actual APS-C film.

645: full frame 645 relative to film is 41.5x56mm for most cameras and 42x56mm for Pentax. The so-called "645" digital sensors are around 40.3x53.7mm.

 

[Erik Kaffehr]

Erik, this is a minor point, but:

APS-C: the common "APS-C" film sensors are 23.5x15.6mm, and the Canon "APS-C" sensors are 22.3x14.9mm. The 25.1x16.7mm dimension I think only applies to actual APS-C film.

645: full frame 645 relative to film is 41.5x56mm for most cameras and 42x56mm for Pentax. The so-called "645" digital sensors are around 40.3x53.7mm.

Absolutely!

Thanks for making that point!

I was pretty aware of being a bit sloppy when posting, but wanted to keep the amount of background checks and text length down.

Thanks for elaborating!

Best regards

Erik

 

[J A C S]

A few remarks:

Lenses are designed for a specific format, so lenses for smaller sensors are sharper physically in general. Then you lose when you magnify. It is a non-linear relation, depending on factors other than sensor size, not easy to quantify.

X-trans is good for certain regular patterns but aliases as much as Bayer overall. There is no free lunch. On the top of that, the demosaicing is harder.

The pattern you used to evaluate aliasing has dominating frequencies.

 

[D Cox]

"If we have a larger sensor, it can detect more photons at the same exposure."

However, it is the exposure that matters -- the number of photons per unit area. Or possibly the number per pixel.

Consider a group of pixels, perhaps ten by ten. Each of these pixels is a light meter, giving a reading. (Ignore colour mosaics for the moment.)

Each reading is subject to uncertainty, because of quantum fluctuations. All physical measurements are subject to uncertainty. We want to know whether the exposure (the number of photons per pixel) is different in different parts of this 10x10 array of measuring instruments.

For example, here is a monochrome image





and here is a 10x10 group of pixels that record the tip of a twig. (There is some corruption from the screen grabbing process).





How do we know that these measurements made in the camera tell us that there was a twig in the scene ? Would adding more pixels around the edge of the whole image make any difference to the reliability of these 100 pixels ? Would making them larger or smaller make a difference ?

The viewed image is probably larger than the sensor (although thumbnails are useful for searching). The greater the enlargement, the easier it is to see the random local variations in the measurements from the sensor. This is where the area of the sensor comes in -- not because at a given exposure the total number of photons is directly proportional to the area, but because the image from a bigger sensor needs less enlargement.

 

[J A C S]

"If we have a larger sensor, it can detect more photons at the same exposure."

However, it is the exposure that matters -- the number of photons per unit area. Or possibly the number per pixel.

I cannot believe that we are still discussing this in 2020.

Unit area of what - the sensor or the final print/image?

 

[Horshack]

For short'ish focal lengths FF MILC wins for me. For longer focal lengths MFT wins.

 

[phcd]

This is an interesting thread comparing sensors of different cameras. The Internet makes it easy to download raw samples to discover similarities and differences.

My personal experience notes the Fuji sensors across the line offer fine color output and are quite sharp at the pixel level. They perform similar to the 5 series cameras offered by Canon. Little output sharpening is needed to make wonderfully crisp prints.

Other sensors (name purposely omitted) are not as sharp at the pixel level and require quite a bit more output sharpening to achieve the same acuity as Fuji and Canon.

Again, it is easy to make your own comparisons. Just download the raw images for evaluation.

 

[NAwlins Contrarian]

My personal experience notes the Fuji sensors across the line offer fine color output and are quite sharp at the pixel level.... Little output sharpening is needed to make wonderfully crisp prints.

Other sensors (name purposely omitted) are not as sharp at the pixel level and require quite a bit more output sharpening to achieve the same acuity as Fuji and Canon.

I think you need to make several distinctions even among Fuji sensors:

* The GFX cameras and some of the low-end of the X-series use Bayer sensors, while the upper end of the X-series uses X-Trans sensors. Their detail resolutions at least, and arguably their sharpnesses, differ.

* The GFX 50S / GFX 50R sensor uses gapped microlenses that increase acutance / apparent sharpness at the expense of somewhat increased noise. AFAIK, the GFX 100 sensor and all of the X-series sensors use full-coverage microlenses.

* AFAIK, some of the cameras have anti-alias filters and some do not.

And of course, the issue of whether or not there's an anti-alias filter, and if so how strong it is, likewise pertains to most brands; even among the Canon, Nikon, Sony, etc. lines there are important differences.

 

[Erik Kaffehr]

My personal experience notes the Fuji sensors across the line offer fine color output and are quite sharp at the pixel level.... Little output sharpening is needed to make wonderfully crisp prints.

Other sensors (name purposely omitted) are not as sharp at the pixel level and require quite a bit more output sharpening to achieve the same acuity as Fuji and Canon.

I think you need to make several distinctions even among Fuji sensors:

* The GFX cameras and some of the low-end of the X-series use Bayer sensors, while the upper end of the X-series uses X-Trans sensors. Their detail resolutions at least, and arguably their sharpnesses, differ.

X-Trans obviously needs a different de-mosaic algorithm than bayer.

* The GFX 50S / GFX 50R sensor uses gapped microlenses that increase acutance / apparent sharpness at the expense of somewhat increased noise. AFAIK, the GFX 100 sensor and all of the X-series sensors use full-coverage microlenses.

I may seem that the difference may be significant. Quite obviously, the GFX 50 sensor has a lot of aliasing.

* AFAIK, some of the cameras have anti-alias filters and some do not.

And of course, the issue of whether or not there's an anti-alias filter, and if so how strong it is, likewise pertains to most brands; even among the Canon, Nikon, Sony, etc. lines there are important differences.

My take was a bit that the Sony A7rIV yielded a much better image quality than the A7III, judged from the studio test shots.

It may also be that a high quality APS-C system may make a lot of sense if we don't print large.

Best regards

Erik

 

[Erik Kaffehr]

Hi,

Doing this comparison I sort of feel that a well designed APS-C system may make more sense than a 24 MP 24x36 mm system that needs an OLP filter to limit color aliasing.

Also, a 24 (or so) MP APS-C may deliver good enough image quality for desktop printer outputs.

Most high quality lenses for 24x36 mm are fast lenses, making them both expensive and heavy.

I would have some hopes that Sigma will make lenses for mirrorless with smaller maximum apertures but still high quality.

Best regards

Erik

 

[J A C S]

Hi,

Doing this comparison I sort of feel that a well designed APS-C system may make more sense than a 24 MP 24x36 mm system that needs an OLP filter to limit color aliasing.

Why? There is nothing wrong with an AA filter.

Also, a 24 (or so) MP APS-C may deliver good enough image quality for desktop printer outputs.

"Good enough" is subjective. How do you get "good enough" f/1.2-f/1.4 shots on APS-C, for example?

Most high quality lenses for 24x36 mm are fast lenses, making them both expensive and heavy.

This applies to APS-C as well; the only problem that they are not as fast (in eq. terms). The main reason I switched to FF was that I was paying a lot for great lenses and getting results worse than FF with the same or cheaper ones.

I would have some hopes that Sigma will make lenses for mirrorless with smaller maximum apertures but still high quality.

Canon will announce an 85/2 and a 50/1.8 soon.

 

[Erik Kaffehr]

Hi,

Doing this comparison I sort of feel that a well designed APS-C system may make more sense than a 24 MP 24x36 mm system that needs an OLP filter to limit color aliasing.

Why? There is nothing wrong with an AA filter.

My argument may be that it is better to have higher resolution on the sensor than discarding resolution. Do proper sampling. Sony now has the same pixel design in at least four sensors in different sizes 26 MP (APS-C), 61 MP (24x36), 102 MP(33x44) and 151 MP(54x41). That makes sense. Now, even high resolution sensors may need OLP filters, I have no issue with that.

Also, a 24 (or so) MP APS-C may deliver good enough image quality for desktop printer outputs.

"Good enough" is subjective. How do you get "good enough" f/1.2-f/1.4 shots on APS-C, for example?

Yes, good enough is subjective, or semi subjective. There is a lot of data on human vision, but there is also a lot of variation in vision.

I don't know how much large apertures are used. They offer very thin DoF. Some folks love ultra thin DoF. Personally I rarely go below f/8 on 24x36 mm or f/11 on MFD. I don't use APS-C.

Most high quality lenses for 24x36 mm are fast lenses, making them both expensive and heavy.

This applies to APS-C as well; the only problem that they are not as fast (in eq. terms). The main reason I switched to FF was that I was paying a lot for great lenses and getting results worse than FF with the same or cheaper ones.

Yes, that is quite possible. Different photographers have different needs.

I would have some hopes that Sigma will make lenses for mirrorless with smaller maximum apertures but still high quality.

Canon will announce an 85/2 and a 50/1.8 soon.

Yes, but I cannot use Canon R-mount lenses on my Sony A7rII.

The other point is that I want lenses that are as good as they get. That means essentially Otus or Sigma Art. But Sigma Art lenses are heavy. I would rather have like the Sigma 45/2.8 lens.

Sony seems to have some pretty decent f/1.8 primes, but they don't count to 'G' or 'GM' series.

Personally, I use Sony A7rII with mostly Canon mount lenses. I would have some interest in moderate aperture primes, like the Voigtlander Apo Lantars or the Zeiss Loxia line, but I don't think the Loxia 35 and 50 lenses are up to date designs.

On the other hand, Canon may go in a direction I like, so I may switch to Canon.

Two friends of mine switched from Canon full frame to Fujifilm X. Don't exactly know why.

Best regards

Erik

 

[J A C S]

Hi,

Doing this comparison I sort of feel that a well designed APS-C system may make more sense than a 24 MP 24x36 mm system that needs an OLP filter to limit color aliasing.

Why? There is nothing wrong with an AA filter.

My argument may be that it is better to have higher resolution on the sensor than discarding resolution. Do proper sampling.

That is what the AA filters do (kinda). The reason resolution is discarded is that there is plenty of it. The APS-C sensors can still benefit from an AA filter, BTW. Also, FF sensors can and sometimes do have a lot of pixels.

Sony now has the same pixel design in at least four sensors in different sizes 26 MP (APS-C), 61 MP (24x36), 102 MP(33x44) and 151 MP(54x41). That makes sense. Now, even high resolution sensors may need OLP filters, I have no issue with that.

Also, a 24 (or so) MP APS-C may deliver good enough image quality for desktop printer outputs.

"Good enough" is subjective. How do you get "good enough" f/1.2-f/1.4 shots on APS-C, for example?

Yes, good enough is subjective, or semi subjective. There is a lot of data on human vision, but there is also a lot of variation in vision.

I don't know how much large apertures are used. They offer very thin DoF. Some folks love ultra thin DoF. Personally I rarely go below f/8 on 24x36 mm or f/11 on MFD. I don't use APS-C.

I do not focus very close at f/1.2-1.4. The DOF is usually quite deep, sometimes even infinite (with my 35/1.4).

The other point is that I want lenses that are as good as they get. That means essentially Otus or Sigma Art. But Sigma Art lenses are heavy. I would rather have like the Sigma 45/2.8 lens.

I am not big fan of the Sigmas. I like that they make heavy lenses, which means fewer compromises. Their vignetting is low, for example. Their AF is iffy though and the bokeh can be distracting.

Sony seems to have some pretty decent f/1.8 primes, but they don't count to 'G' or 'GM' series.

Personally, I use Sony A7rII with mostly Canon mount lenses. I would have some interest in moderate aperture primes, like the Voigtlander Apo Lantars or the Zeiss Loxia line, but I don't think the Loxia 35 and 50 lenses are up to date designs.

On the other hand, Canon may go in a direction I like, so I may switch to Canon.

Two friends of mine switched from Canon full frame to Fujifilm X. Don't exactly know why.

This is a good example for expensive crop lenses. They are very good, indeed but they do not save you money.

 

[Erik Kaffehr]

I was shooting nine years ago with a 24 MP (24x36) Sony A900. At the same time I also had a Sony A55 SLT, 16MP APS-C. The SLT was not one of my favourites but it allowed focusing by magnified live view. It was also using a newer generation sensor.

Found a shooting position with a good view on some autumn vegetation. Problem was that it was a bit darkish and it was windy. So, I got quite concerned about motion blur from the wind. So, I used both cameras.

With the A900 I used the Sony 70-300/4.5-5.6 G at f/5.6 while with the A55 I would use the Zeiss labeled Sony 24-70/2.8 at f/4. In the end it did not help with shutter speeds. But, the A55/SLT I could focus with more confidence.

So, I printed both at A2 and making the choice was hard. In the end I hung the 16 MP APS-C image on the wall.



[ATTACH alt="This shows A2 size at 180 PPI, a good way to judge print quality on a 24" 1920x1200 monitor at actual pixels, at least in my humble experience. "]2515459[/ATTACH]
This shows A2 size at 180 PPI, a good way to judge print quality on a 24" 1920x1200 monitor at actual pixels, at least in my humble experience.

A55 (16MP APS-C)

A900 (24MP FF)

Best regards

Erik

--
Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic uses to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

 

[Erik Kaffehr]

For short'ish focal lengths FF MILC wins for me. For longer focal lengths MFT wins.

I am pretty sure I see what you mean.

Best regards

Erik

 

[57even]

Hi,

Doing this comparison I sort of feel that a well designed APS-C system may make more sense than a 24 MP 24x36 mm system that needs an OLP filter to limit color aliasing.

That was my conclusion. I traded all my Nikon gear on an Xpro2 after my experience with an Xpro1. Xpro1 was better than D700, and Xpro2 proved to be better than a D600. Actually, pretty close to my D800 (with AA filter) which I also traded.

Also, a 24 (or so) MP APS-C may deliver good enough image quality for desktop printer outputs.

Will print borderless 24X16 at 250 PPI and still looks great close up. I have to get them done professionally as my desktop doesn't exceed 19X13.

Usually printed on A1, which gives a nice border, and mounted on MTF.

However, Xtrans demosaicing needs care with some subjects.

Most high quality lenses for 24x36 mm are fast lenses, making them both expensive and heavy.

This is very true. With Fuji, I have both options, heavy and fast or light and sharp.

I would have some hopes that Sigma will make lenses for mirrorless with smaller maximum apertures but still high quality.

Not felt the lack so far. As you suggested, 24 MP APSC with no AA filter is somewhat sharper than 24 MP FF with one, with APSC lenses.

 

[rbf]

A few remarks:

Lenses are designed for a specific format, so lenses for smaller sensors are sharper physically in general. Then you lose when you magnify. It is a non-linear relation, depending on factors other than sensor size, not easy to quantify.

X-trans is good for certain regular patterns but aliases as much as Bayer overall. There is no free lunch. On the top of that, the demosaicing is harder.

The pattern you used to evaluate aliasing has dominating frequencies.

I see this said often but never have seen any visual evidence of aliasing from an XTrans sensor. I've had them since I got my XE1. I've also noticed that cell phone cameras have been exploring alternative CFA's for a while and getting good reviews. Dr Bayer of Kodak came up with a good algorithm that has been widely used for decades. But there appears to room for innovation to me.

 

[Erik Kaffehr]

Comparing the 26MP Fujifilm X-T4 with the Sony A7III the APS-C camera has a small advantage in the 'worst direction' on the A7III,




The GFX 50S has a small advantage over the A7rIV, probably due to reduced pixel aperture in combination with a better lens.






Comparing the GFX 100 with the GFX 50S using DPReviews Studio Shots indicate that GFX 100 delivers much more detail. Also, both have similar MTF at Nyquist, although the GFX 50S shows much more aliasing in the image. Jim Kasson found that the MTF for the GFX 50S and GFX 100 were essentially similar. So, here the data goes apart, which is a bit confusing. Jim's data is in agreement with what I see, however.

Best regards

Erik

--
Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic uses to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

 

[Jack Hogan]

For short'ish focal lengths FF MILC wins for me. For longer focal lengths MFT wins.

How do you figure that for longer focal lengths Horshack?

My birding friends tell me that "there's no substitute for square mm of pupil area". If that premise is true, equivalent lenses are going to end up pretty well having similar bulk/weight/cost on all real photography formats (some a bit better, some a bit worse but in the ballpark). If they don't, buyer beware because it's likely that the relative manufacturer has taken shortcuts, hoping that its intended audience will not notice that they are no longer comparing apples to apples vs the competition

Jack