Pixel size more important than megapixel amount for image quality?

[Peter Mathews]

The larger pixels of a lower resolution sensor have less noise (the first place) therefore require less NR. What's the point of higher resolution if the individual pixel signals are so noisy you have to apply aggressive NR to get rid of it thus, smudging detail.

Peter,

Why do you think that the noise of an image is proportional to the noise of its individual pixels?

If you had two cameras with sensors that measured 3mm x 2mm, neither camera added any noise, and one of the cameras had 6 pixels and the other had 24 pixels, what would be the ratio of the SNR of a pixel on the 6 pixel sensor to that of a pixel on the 24MP sensor at the same exposure? What would be the ratio of the SNR of the image from the 6 pixel sensor to the SNR of the image of the 24MP sensor.

Yes, theoretically, in a perfect system, the size of the pixels wouldn't matter as long as they covered the same surface area however, there are losses due to physical problems. The space in between the pixels for example. The space between can't gather photons so there is a loss if you use 4 pixels instead of one for any given area. It may be small but, it's there. Also, larger pixels can be made more efficient (conductor dimensions and amplifier efficiency just to name a couple). There is also the amplification factor. 1/4 the size means twice the gain. More gain equals more noise.

If the theory that more pixels for any given surface area is more critical than actual pixel dimensions than the 48MP A7R-II should literally blow the A7S-II out of the water in image quality at higher ISO's but it doesn't. At best, it comes close but empirical evidence (forget DxoMark rubbish) suggests that lower resolution sensors at current technology levels still hold the upper hand. Why else would Sony make the A7S-II still?

Studio comparison scenes where a smaller resolution image is upscaled (or cropped) to match the higher resolution image are obfuscations because cropping is akin to a signal amplification so if the A7S-II image cropped to match the A7R-II image shows the exact same noise, you have to conclude the inherent SNR in the A7R-II image is greater than the A7S-II image by whatever the crop factor used to get the images displayed to the same size. So, if we take a picture of a sphere at 50 meters with both cameras and enlarge the image from the 12MP sensor so that the size of the sphere matches the image from the 48MP sensor and the noise is the same, the larger sensor image has to have a greater SNR!

This is from Cambridgeincoulour.com-

"Does all this mean it is bad to squeeze more pixels into the same sensor area? This will usually produce more noise, but only when viewed at 100% on your computer monitor. In an actual print, the higher megapixel model's noise will be much more finely spaced — even though it appears noisier on screen (see "Image Noise: Frequency and Magnitude"). This advantage usually offsets any increase in noise when going to a larger megapixel model (with a few exceptions)."

So, the small pixel, higher resolution sensor "seems" to have the same SNR in actual print. It's when you start cropping it becomes clear it's not.

That's my whole point. If you get the same "perceived" SNR from a 48MP sensor camera than from a 12MP camera? Why spend more money on the more expensive sensor? Any advantage in resolution is lost due to the need to reduce the greater noise, even at base ISO.

I was teetering on getting the D810 vs. my D750 now I'm glad I didn't.

 

[Peter Mathews]

I am trying to understand what their roles are when it comes to image quality.

thanks

 

[Peter Mathews]

bigger pixel, more light- as simple as that

This shows a lack of understanding of how to apply this fact. Please read Great Bustard's post above -> carefully.

Your statement may also indicate a a definition of image quality that is too narrow for the venue. There are many needs that may have to be served by the same sensor.

 

[FingerPainter]

The larger pixels of a lower resolution sensor have less noise (the first place) therefore require less NR. What's the point of higher resolution if the individual pixel signals are so noisy you have to apply aggressive NR to get rid of it thus, smudging detail.

Peter,

Why do you think that the noise of an image is proportional to the noise of its individual pixels?

If you had two cameras with sensors that measured 3mm x 2mm, neither camera added any noise, and one of the cameras had 6 pixels and the other had 24 pixels, what would be the ratio of the SNR of a pixel on the 6 pixel sensor to that of a pixel on the 24MP sensor at the same exposure? What would be the ratio of the SNR of the image from the 6 pixel sensor to the SNR of the image of the 24MP sensor.

Yes, theoretically, in a perfect system, the size of the pixels wouldn't matter as long as they covered the same surface area however, there are losses due to physical problems. The space in between the pixels for example. The space between can't gather photons so there is a loss if you use 4 pixels instead of one for any given area. It may be small but, it's there.

Given the size of the pixels in our little example it is quite reasonable to assert that space between them is scaled so the same amount of area is wasted per sensor. TO cover any other factors you may wish to being up, I'll just specify that the two sensors have the same quantum efficiency. So please stop ducking the question.

Also, larger pixels can be made more efficient (conductor dimensions and amplifier efficiency just to name a couple).

Now covered by the assertion of same same QE,

There is also the amplification factor. 1/4 the size means twice the gain.

Why not four times the gain?

More gain equals more noise.

Yes but not a higher SNR, because the signal is gained. If the gain applied to voltage from sensors in digital cameras resulted in a larger increase in noise than in signal, we'd see different noise performance at high ISOs than what we actually find in tests. What we actually see is that if we increase gain by raising ISO, but keep exposure unchanged, the SNR increases, not decreases. That's because adding gain doesn't increase noise more than signal. Adding gain increases al the signal by the factor of the gain but only increases the noise that is in the light or added by the camera up to the point where gain is applied, but doesn't increase noise added after gain is applied.

If the theory that more pixels for any given surface area is more critical than actual pixel dimensions than the 48MP A7R-II should literally blow the A7S-II out of the water in image quality at higher ISO's but it doesn't.

Your conclusion doesn't follow from the premise. You also seem to be sloppily exchanging different aspects of IQ.

At best, it comes close but empirical evidence (forget DxoMark rubbish) suggests that lower resolution sensors at current technology levels still hold the upper hand. Why else would Sony make the A7S-II still?

Fast read-out for video.

<snip stuff irrelevant to the questions I asked you>

This is from Cambridgeincoulour.com-

"Does all this mean it is bad to squeeze more pixels into the same sensor area? This will usually produce more noise, but only when viewed at 100% on your computer monitor.

IOW, the pixel level noise of a smaller pixel is higher. No surprise there, but not relevant to how a whole image looks.

In an actual print, the higher megapixel model's noise will be much more finely spaced — even though it appears noisier on screen (see "Image Noise: Frequency and Magnitude"). This advantage usually offsets any increase in noise when going to a larger megapixel model (with a few exceptions)."

So, the small pixel, higher resolution sensor "seems" to have the same SNR in actual print. It's when you start cropping it becomes clear it's not.

No. there is a difference between the SNR of an image and the SNR of its individual pixels. I asked you the questions about the 6 and 24MP sensors to see if you could demonstrate that you understood those differences. SIince you have avoided answering, it would be easy to presume you don't understand.

That's my whole point. If you get the same "perceived" SNR from a 48MP sensor camera than from a 12MP camera? Why spend more money on the more expensive sensor?

Because you also get a much higher perceived sharpness with the 48MP sensor. Result: higher IQ.

Any advantage in resolution is lost due to the need to reduce the greater noise, even at base ISO.

No it is not. If we made a 30" x 20" print from the 48MP sensor and from the 12MP sensors, without applying any noise reduction to either image, the print from the 48MP would look appreciably sharper but there would be no noticeable difference in noise when viewed from the same distance.

I was teetering on getting the D810 vs. my D750 now I'm glad I didn't.

It's a pity you are making decisions based on failed understanding.

 

[Peter Mathews]

The larger pixels of a lower resolution sensor have less noise (the first place) therefore require less NR. What's the point of higher resolution if the individual pixel signals are so noisy you have to apply aggressive NR to get rid of it thus, smudging detail.

Peter,

Why do you think that the noise of an image is proportional to the noise of its individual pixels?

If you had two cameras with sensors that measured 3mm x 2mm, neither camera added any noise, and one of the cameras had 6 pixels and the other had 24 pixels, what would be the ratio of the SNR of a pixel on the 6 pixel sensor to that of a pixel on the 24MP sensor at the same exposure? What would be the ratio of the SNR of the image from the 6 pixel sensor to the SNR of the image of the 24MP sensor.

Yes, theoretically, in a perfect system, the size of the pixels wouldn't matter as long as they covered the same surface area however, there are losses due to physical problems. The space in between the pixels for example. The space between can't gather photons so there is a loss if you use 4 pixels instead of one for any given area. It may be small but, it's there.

Given the size of the pixels in our little example it is quite reasonable to assert that space between them is scaled so the same amount of area is wasted per sensor. TO cover any other factors you may wish to being up, I'll just specify that the two sensors have the same quantum efficiency. So please stop ducking the question.

Also, larger pixels can be made more efficient (conductor dimensions and amplifier efficiency just to name a couple).

Now covered by the assertion of same same QE,

There is also the amplification factor. 1/4 the size means twice the gain.

Why not four times the gain?

More gain equals more noise.

Yes but not a higher SNR, because the signal is gained. If the gain applied to voltage from sensors in digital cameras resulted in a larger increase in noise than in signal, we'd see different noise performance at high ISOs than what we actually find in tests. What we actually see is that if we increase gain by raising ISO, but keep exposure unchanged, the SNR increases, not decreases. That's because adding gain doesn't increase noise more than signal. Adding gain increases al the signal by the factor of the gain but only increases the noise that is in the light or added by the camera up to the point where gain is applied, but doesn't increase noise added after gain is applied.

If the theory that more pixels for any given surface area is more critical than actual pixel dimensions than the 48MP A7R-II should literally blow the A7S-II out of the water in image quality at higher ISO's but it doesn't.

Your conclusion doesn't follow from the premise. You also seem to be sloppily exchanging different aspects of IQ.

At best, it comes close but empirical evidence (forget DxoMark rubbish) suggests that lower resolution sensors at current technology levels still hold the upper hand. Why else would Sony make the A7S-II still?

Fast read-out for video.

<snip stuff irrelevant to the questions I asked you>

This is from Cambridgeincoulour.com-

"Does all this mean it is bad to squeeze more pixels into the same sensor area? This will usually produce more noise, but only when viewed at 100% on your computer monitor.

IOW, the pixel level noise of a smaller pixel is higher. No surprise there, but not relevant to how a whole image looks.

In an actual print, the higher megapixel model's noise will be much more finely spaced — even though it appears noisier on screen (see "Image Noise: Frequency and Magnitude"). This advantage usually offsets any increase in noise when going to a larger megapixel model (with a few exceptions)."

So, the small pixel, higher resolution sensor "seems" to have the same SNR in actual print. It's when you start cropping it becomes clear it's not.

No. there is a difference between the SNR of an image and the SNR of its individual pixels. I asked you the questions about the 6 and 24MP sensors to see if you could demonstrate that you understood those differences. SIince you have avoided answering, it would be easy to presume you don't understand.

That's my whole point. If you get the same "perceived" SNR from a 48MP sensor camera than from a 12MP camera? Why spend more money on the more expensive sensor?

Because you also get a much higher perceived sharpness with the 48MP sensor. Result: higher IQ.

Any advantage in resolution is lost due to the need to reduce the greater noise, even at base ISO.

No it is not. If we made a 30" x 20" print from the 48MP sensor and from the 12MP sensors, without applying any noise reduction to either image, the print from the 48MP would look appreciably sharper but there would be no noticeable difference in noise when viewed from the same distance.

I was teetering on getting the D810 vs. my D750 now I'm glad I didn't.

It's a pity you are making decisions based on failed understanding.

Hey, your welcome to your opinion. I have degree in electronics and 40 years in the profession. I know a little about SNR.

Good luck with your beliefs.

 

[Julia Borg]

There is also the amplification factor. 1/4 the size means twice the gain.

Why not four times the gain?

what your opponent does not know is that the higher the gain is the better is SNR.

he may not realize that a shallow well is an advantage exactly because the gain can be higher.

oh well. he may not even know what gain he is talking about. he may be even thinking ISO.

--
Julia

 

[Mark Blauhoefer]

Dang it - you're right! I reduced the 200 px across to 40 px across, and then got confussed when I enlarged it

 

[FingerPainter]

There is also the amplification factor. 1/4 the size means twice the gain.

Why not four times the gain?

what your opponent does not know is that the higher the gain is the better is SNR.

he may not realize that a shallow well is an advantage exactly because the gain can be higher.

oh well. he may not even know what gain he is talking about. he may be even thinking ISO.

Indeed.

But he seems to think that his degree in electronics and 40 years in the profession tell him otherwise.

 

[Julia Borg]

There is also the amplification factor. 1/4 the size means twice the gain.

Why not four times the gain?

what your opponent does not know is that the higher the gain is the better is SNR.

he may not realize that a shallow well is an advantage exactly because the gain can be higher.

oh well. he may not even know what gain he is talking about. he may be even thinking ISO.

Indeed.

But he seems to think that his degree in electronics and 40 years in the profession tell him otherwise.

a man with a degree is born every minute.

 

[Great Bustard]

5years ago pixel size was important. Now its the opposite ? Go figure all the tech heads have no idea.

As usual, you are mistaken. As always, the pixel size matters only inasmuch as it contributes to the total number of pixels that make up the photo. By the way, didn't know you were a girl:

;-)

Would you like to share with us all why the pen f has equal image quality at I so 12000 to the d5 ? Going by your logic ;-) both 20meg sensors.

I ask because I'd be a bit concerned if you considered this to represent "equal image quality":





Above we see mFT, FF, and APS-C at the same exposure at ISO 12800. Pretty easy to figure out which is which even if they weren't labeled, really.

 

[J A C S]

There is also the amplification factor. 1/4 the size means twice the gain.

Why not four times the gain?

what your opponent does not know is that the higher the gain is the better is SNR.

he may not realize that a shallow well is an advantage exactly because the gain can be higher.

oh well. he may not even know what gain he is talking about. he may be even thinking ISO.

Indeed.

But he seems to think that his degree in electronics and 40 years in the profession tell him otherwise.

It is funny that somebody with a degree would think that 1/4 of the area would have to be brightened.

 

[Great Bustard]

I am trying to understand what their roles are when it comes to image quality.

thanks

It's not the size of the pixel that matters, but the number of pixels that matters, and the total amount of light that all those pixels record.

Consider a 12 MP and 48 MP sensor, both with the same size. Four of the pixels on the 48 MP sensor will record the same proportion of the scene as one of the pixels from the 12 MP sensor, and also record the same amount of light. The difference is that the four pixels will record more detail than one pixel.

Now, there is another consideration: the overall noise in the photo. Most of the noise in the photo comes from the amount of light that makes up the photo -- the more light that makes up the photo, the less noisy it is -- but as mentioned above, the pixel count doesn't influence the amount of light that makes up the photo (as much as some may wish to erroneously argue otherwise). However, the sensor and supporting hardware also add in noise of their own (electronic noise), and more smaller pixels are typically more noisy than fewer larger pixels, all else equal.

On the other hand, this electronic noise plays a very small role until the light gets very low. For example, we might not notice any difference at all until we get to very high ISO settings, like ISO 6400 (FF equivalent).

On the other hand, the greater resolution of the 48 MP photo allows more aggressive noise filtering to be used. Thus, after noise filtering is judiciously applied, the 48 MP photo may well be just as noisy as the 12 MP photo but retaining more detail, or less noisy for the same amount of detail (depending on how aggressive you are with the noise filtering).

See here for a more thorough explanation and here for some excellent examples of what I'm talking about.

If that's true, shouldn't the SNR of the A7RII be significantly higher than the A7SII?

No, they should be about the same, and they are. For example:



At the same display size and ISO 1600 (top row), they are both essentially equally noisy. At the same display size and 1/4 the light (ISO 6400 -- bottom row), we see the A7S2 has a very slight noise advantage. However, that's simply by downsampling. If, instead, we applied noise filtering to the A7R2 photo first, then the differences would be significantly less, most likely favoring the A7R2.

That's right. At the same "display size". The lower resolution but larger pixel sensor output has to be cropped to match display size. The net effect is an amplification of the grain so the perceived noise level is equal, which it's not on the pixel level. The A7SII test sample would be noticeably less grainy if magnified optically.

So what's the point of a 48MP sensor if more aggressive NR is required to get the same IQ as a 12MP sensor at higher ISO's?

Not much, if any, in very low light. But a rather substantial advantage in good light.

So there isn't, you admit. Yes, in good light, higher resolution is an advantage but at what display size. Only past a certain size where 300dpi can no longer be maintained by the lower resolution sensor.

In all cases, I'm talking about the same display size. The fewer larger pixels only come out ahead in very low light if noise filtering is not used on the greater number of smaller pixels.

After all, is the final output of each sensor relatively the same after aggressive NR is applied to the higher resolution sensor and would the smaller pixel of the 48MP senor not be more prone to shot noise and saturation? Aren't you basically paying for a 48MP sensor and getting a 12MP result at ISO's above 800?

The point is that more pixels give you more IQ options than fewer pixels. It's akin to asking what's the advantage of RAW over the default OOC jpg -- for some purposes, no advantage; for others, a rather clear advantage, and other times, somewhere in between.

I disagree. Primarily in that image quality is highly subjective and high resolution sensors as you stated, have their envelopes of operational advantage but overall, on balance, the net gains are offset by deficiencies. Certainly for landscapes with high detail, in good light, I would pick the 48MP sensor (A7RII) hands down. For low light work or where fine detail is not critical (as is much the case in general), the 12MP sensor seems more appropriate and does have a distinct advantage.

But the A7S2 does not have a "distinct advantage" over the A7R2 even for low light (well, at least up to ISO 25600 -- maybe beyond that it does).

Each type of sensor has it's own characteristics and higher resolution is no panacea for ailments in image quality at lower signal levels. That's the precise reason Sony makes an A7SII and it's well received for it's prowess.

No, Sony made the A7S2 for video, since they were unable to bin the 42 MP of the A7R2 to 4K.

If I need to downsample the 48MP image to get the same level of noise as the 12MP, thus losing the resolution advantage, there is no objective justification for using the term "Quite the opposite" because it's not so quite, and not so opposite.

You are mistaken. The greater number of smaller pixels has a resolution advantage with the same noise in good light. As the light dims, the resolution advantage gets eaten into as noise filtering is used more and more to compensate for the noise. But the fewer number of larger pixels do not come out ahead (except, perhaps, in very, very, very low light, such as ISO 51200+, because even at ISO 25600, the evidence doesn't show an advantage for the A7S2).

It's not your answer that's flawed it's the question. Image quality under what circumstance would be more appropriate.

If someone was interested in the "highest IQ" for stills and could only choose between the A7S2 and the A7R2, the A7R2 would be the winner. In good light, it would be the winner in a big way. As the light gets lower and lower, the margin of victory gets more and more narrow and, perhaps, reverses in *extremely* low light (at least ISO 51200+, but I'm not certain it ever reverses).

 

[Great Bustard]

If that's true, shouldn't the SNR of the A7RII be significantly higher than the A7SII?

Funny you mention that bc it actually is as good or better at nearly all ISOs. Check DXO and compare both, the A7rII is better at ISO 100-1600, is the same at 3200-25k, takes the lead again at ISO 51k, and only falls behind at it's max of 100k.

This is in line with what has been explained, that when photons get scarce in dim light (when ISO climbs), there is less signal to cover up the noise. That is when read noise is more noticeable. Other than ISO 100k and beyond, the A7sII has no advantage.

On top of this the 42mp sensor has far better color scores at nearly every ISO stop (they are even at ISO 100k). Now is it possible Sony just "made a better sensor" in the 42mp one? Sure that's possible, but it shows that pixel counts make less difference than so many other factors you shouldnt be basing your choice on which has less.

I don't think you get it. Apart from DXoMark being garbage data...

Um...

...the claim was made that it's "quite the opposite". If it's the same. it's not the opposite so the statement is false.

It is, in fact, quite the opposite.

The theory is that you now need 4 pixels instead of one to get the same SNR which defeats the purpose of the higher resolution sensor, at least in low light.

In *very* low light.

Then look at the color noise in the output of the A7RII at 12800 as compared to the A7SII and you'll see where the breakdown lies. 12.8K is the new benchmark for ISO performance as most modern FF camera do quite well up to 6400 nowadays.

You might be surprised to learn that people, as a whole, shoot below ISO 12800 more often than above ISO 12800. That said, do have a look at the A7R2 vs the A7S2 at ISO 25600. Not seeing an IQ advantage to the fewer larger pixels even then.

Of course, if you're always shooting ISO 12800+, then, sure, what's the point of the A7R2 over the A7S2? Surprisingly, most people spend most of their time taking photos in much better light where the advantages of the A7R2 will be significantly more apparent, only to be lost again when viewed on a cell phone. ;-)

 

[Great Bustard]

Girl wants pizza sliced into eight pieces instead of twelve:

... I refuse to believe anybody can be that stupid.

Then there are a lot of actors on DPR. ;-)

mostly men, mind you.

Of course -- male actors get paid more. ;-)

 

[Julia Borg]

Girl wants pizza sliced into eight pieces instead of twelve:

... I refuse to believe anybody can be that stupid.

Then there are a lot of actors on DPR. ;-)

mostly men, mind you.

Of course -- male actors get paid more. ;-)

actually... you don't want to know

 

[Donald B]

The problem GB is half of my posts have been deleted ? So any conclusions you have come up with are way out of context. Any way ir comparrision of the d4s and the d810 at I so 6400 the larger pixels are clearly in front because the d810 couldn't even capture any hair detail . can't use nr or sharpening if their isn't any detail to begin with.

Cheers don

 

[Donald B]

12800

 

[Great Bustard]

The problem GB is half of my posts have been deleted ? So any conclusions you have come up with are way out of context.

Um, Don? I was replying to your post that was a direct reply to mine. Would be pretty weird if you had posts in between, no?

Any way ir comparrision of the d4s and the d810 at I so 6400 the larger pixels are clearly in front because the d810 couldn't even capture any hair detail . can't use nr or sharpening if their isn't any detail to begin with.

Don, as I linked to at the end of my entry to this thread, here's the D4s vs the D810 at ISO 6400. And, well, it should come as no surprise that it demonstrates exactly what I was saying.

 

[Great Bustard]

Not much, if any, in very low light. But a rather substantial advantage in good light.

So there isn't, you admit. Yes, in good light, higher resolution is an advantage but at what display size. Only past a certain size where 300dpi can no longer be maintained by the lower resolution sensor.

The advantages to higher megapixels are not limited to final display size. More megapixels allow much more headroom for all sorts of processes applied before downsampling, and can be quite noticeable even on Instagram:

• cropping (obviously?)
• rotation
• perspective correction
• distortion correction
• more aggressive luminance and chroma NR
• edge detection/clarity
• sharpness

Who's questioning high resolution? I though the question was does high res trump low resolution for technical image quality. This seems to be clearly not the case in many instances. Based on the evidence at hand, In my view anyway.

The evidence at hand shows that, even in very, very, very low light, more smaller pixels are no worse for the wear:

• D4s vs D810 at ISO 6400
• A7R2 vs A7S2 at ISO 25600

Funny thing is, people are below ISO 6400 a lot more than they are above it. Perhaps you're one of the rare exceptions, though.

Why is pixel density still in dispute. This is ridiculous.

Agreed, what with evidence like the above plain for all to see, and all.

 

[Max Iso]

The problem GB is half of my posts have been deleted ? So any conclusions you have come up with are way out of context. Any way ir comparrision of the d4s and the d810 at I so 6400 the larger pixels are clearly in front because the d810 couldn't even capture any hair detail . can't use nr or sharpening if their isn't any detail to begin with.

Cheers don