This article should be read by everybody

Started Dec 12, 2013 | Discussions
Shop cameras & lenses ▾
tbcass
tbcass Forum Pro • Posts: 24,941
This article should be read by everybody
6

Many people will refuse to acknowledge it's validity because it goes against some misconceptions that some people incorrectly believe are truths.

http://www.luminous-landscape.com/essays/dxomark_sensor_for_benchmarking_cameras2.shtml

The article is lengthy, written by a Physicist and is backed up by mathematical proofs. One of his statements from the article:

" Resolution doesn’t increase noise

Increasing resolution (MPixels) for a given sensor size has no direct impact on image noise. In fact, some of the lowest noise cameras (Nikon D800, Nikon D3x, Pentax K-5, Nikon D7000, Sony Alpha 580) have relatively high resolutions. See Figure 1a."

-- hide signature --

Tom
Look at the picture, not the pixels
------------
Miss use of the ability to do 100% pixel peeping is the bane of digital photography because it causes people to fret over inconsequential issues.
http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/

 tbcass's gear list:tbcass's gear list
Sony RX100 Sony SLT-A77 Tamron SP AF 90mm F/2.8 Di Macro Sony DT 35mm F1.8 SAM Tamron SP 70-300mm F/4-5.6 Di USD +8 more
Nikon D3X Nikon D7000 Nikon D800 Pentax K-5
If you believe there are incorrect tags, please send us this post using our feedback form.
Vikas Rana
Vikas Rana Contributing Member • Posts: 916
Re: This article should be read by everybody

That is a useful information, thanks for sharing Tom!

 Vikas Rana's gear list:Vikas Rana's gear list
Sony SLT-A58 Tamron SP AF 17-50mm F/2.8 XR Di II LD Aspherical (IF) Sony DT 55-300mm F4.5-5.6 SAM Sony DT 18-55mm F3.5-5.6 SAM II Adobe Photoshop CS6
Renato1 Veteran Member • Posts: 3,324
Re: This article should be read by everybody

Interesting read thanks, though I think he was a bit confused early on in the piece saying little cameras like the NEX have nearly as good image quality as a DSLR - it should be as good, as it has the same sized sensor.

The article kept reminded me very much of an article I read in a magazine nearly 20 years ago, where they were demonstrating that ISO400 film was now nearly as good as the ISO100 film of seven or eight years earlier, both in terms of resolution and grain. Same thing is obviously happening with the sensors that replaced film.

Regards,

Renato

PeteC21
PeteC21 Contributing Member • Posts: 705
Re: This article should be read by everybody

Thanks, nice read

-- hide signature --

"If I knew how to take a good photograph, I'd do it every time." - Robert Doisneau

 PeteC21's gear list:PeteC21's gear list
Sony Alpha 7R II Sony Alpha a6300 Sony E 10-18mm F4 OSS Carl Zeiss Touit 1.8/32 Sony Vario-Tessar T* E 16-70mm F4 ZA OSS +6 more
Nordstjernen
Nordstjernen Veteran Member • Posts: 6,876
Hard to bust the large vs small pixel myth
1

tbcass wrote:

" Resolution doesn’t increase noise

Increasing resolution (MPixels) for a given sensor size has no direct impact on image noise. In fact, some of the lowest noise cameras (Nikon D800, Nikon D3x, Pentax K-5, Nikon D7000, Sony Alpha 580) have relatively high resolutions. See Figure 1a."

I think the myth that large pixels gives a cleaner image for sensors of a given size will live on for years. My question is: Why trade resolution for nothing?

There will still be a lot of posters crying for lower res sensors, thinking that insane high ISO settings that is never used for real life photography would look cleaner!

 Nordstjernen's gear list:Nordstjernen's gear list
Sony Alpha a99 Sony Alpha 7
tex Veteran Member • Posts: 6,872
Whew! Well, that should put some things to rest in 2 ways

Ans exhaustive, well reasoned and well documented article.

But for a bit more oomph for the rest of us, do go to his website and visit his gallery(s) to see his images. Many are quite fine.  For me that is a major takeaway:  the interest and understanding of the tech/science does not supersede this guy's eye and aesthetic vision.

I'm going to put it down to his being Dutch, a well balanced group of people generally.  And for a such small nation, one with an outsize contribution to western visual art.

-- hide signature --

tex_andrews, co-founder and webmaster of The LightZone Project, an all-volunteer group providing the free and open source LightZone photo editing software.
"Photography is the product of complete alienation" Marcel Proust
"I would like to see photography make people despise painting until something else will make photography unbearable." Marcel Duchamp

 tex's gear list:tex's gear list
Sony Alpha 7R Pentax 645Z Tamron AF 28-75mm F/2.8 XR Di LD Aspherical (IF) Rokinon T-S 24mm F3.5 ED AS UMC +15 more
K E Hoffman
K E Hoffman Senior Member • Posts: 4,841
Re: Hard to bust the large vs small pixel myth
5

Nordstjernen wrote:

tbcass wrote:

" Resolution doesn’t increase noise

Increasing resolution (MPixels) for a given sensor size has no direct impact on image noise. In fact, some of the lowest noise cameras (Nikon D800, Nikon D3x, Pentax K-5, Nikon D7000, Sony Alpha 580) have relatively high resolutions. See Figure 1a."

I think the myth that large pixels gives a cleaner image for sensors of a given size will live on for years. My question is: Why trade resolution for nothing?

There will still be a lot of posters crying for lower res sensors, thinking that insane high ISO settings that is never used for real life photography would look cleaner!

Physics is still physics.. less light to vs base system noise on smaller pixel more noise is seen.

We are getting to the place where the pixel count is not adding to the image because most of the lenses can't handle it.

Even the charts all bear it out in the article.  The APC sensors are noisier than the FF and you will find that the FF tend to have larger pixels if it didn't matter the measure would be the same for a 25 MP A99 and A77.. its not.

What the article almost obfuscates (it mentions it. then dismisses it) is that the fact that the rule is "broken" so often has to do with all the the other variables.. New Sensor Technology and camera design.. etc that allows an A77 to have a better noise rating that the A700 did.  Would not happen if it were the same generation sensor but double the pixel count.

And yes Printing / scaling creates system that averages out noise in most cases..

But a 16 MP A77 with the same sensor technology would have less noise..  I think we have seen that in the sony line.

If you print to the same size less than 11x14 etc does it tend to smooth out.. yes. as you make larger prints or CROP the noise will not be smoothed out.

I also enjoyed the chart where is said 20 MP cell phone camera couldn't happen - Tell that to Nokia with the Lumia 1020 at 41 MP.  NEVER SAY NEVER related to technology.. odds are you will get burned.

Yes you can't assume a higher res sensor will have more noise than another lower res camera or the previous model.. but it would have more than a sensor with the same design and larger photo reception sites.

-- hide signature --

K.E.H. >> Shooting between raindrops in WA<<

 K E Hoffman's gear list:K E Hoffman's gear list
Canon EOS 450D Nikon 1 J1 Sony Alpha a6300 Sony 100mm F2.8 Macro Sony E 18-200mm F3.5-6.3 OSS +4 more
Slideshow Bob Senior Member • Posts: 1,584
The only problem is that...
6

…it actually doesn't appear to work like that. Take Nikon's current FX lineup as an example, and you'll find that when listed in order of noise levels (best to worst) at the image level (so per area rather than per pixel), you'd get this...

Df

D4

D610

D800

…which is actually proving that for a given sensor size, smaller pixels are actually more noisy. You can't resize a D800 image to DF size and achieve the same (or less) noise without applying NR. NR is a post processing step, whether it's done in camera or on a PC.

Listing Canon FF cameras in the same fashion produces this...

1Dx

6D

5DIII

… with the highest resolution sensor again producing the most noise, and the lowest resolution one producing the least.

I don't doubt the qualifications of the people who wrote that article, but unfortunately, their conclusion just doesn't match what happens in the real world. For a given level of technology, bigger pixels produce less noise.

tbcass wrote:

Many people will refuse to acknowledge it's validity because it goes against some misconceptions that some people incorrectly believe are truths.

http://www.luminous-landscape.com/essays/dxomark_sensor_for_benchmarking_cameras2.shtml

The article is lengthy, written by a Physicist and is backed up by mathematical proofs. One of his statements from the article:

" Resolution doesn’t increase noise

Increasing resolution (MPixels) for a given sensor size has no direct impact on image noise. In fact, some of the lowest noise cameras (Nikon D800, Nikon D3x, Pentax K-5, Nikon D7000, Sony Alpha 580) have relatively high resolutions. See Figure 1a."

Why is it that when people say thing like "…some of the lowest noise cameras…" they don't actually list ANY of the lowest noise cameras, like the 1Dx, D3s, or D4?

SB

TrojMacReady
TrojMacReady Veteran Member • Posts: 8,696
Re: Hard to bust the large vs small pixel myth
3

K E Hoffman wrote:

Nordstjernen wrote:

tbcass wrote:

" Resolution doesn’t increase noise

Increasing resolution (MPixels) for a given sensor size has no direct impact on image noise. In fact, some of the lowest noise cameras (Nikon D800, Nikon D3x, Pentax K-5, Nikon D7000, Sony Alpha 580) have relatively high resolutions. See Figure 1a."

I think the myth that large pixels gives a cleaner image for sensors of a given size will live on for years. My question is: Why trade resolution for nothing?

There will still be a lot of posters crying for lower res sensors, thinking that insane high ISO settings that is never used for real life photography would look cleaner!

Physics is still physics.. less light to vs base system noise on smaller pixel more noise is seen.

Sensor size is the important determinator (plus sensor technology of course) at a given f stop and shutterspeed.

We are getting to the place where the pixel count is not adding to the image because most of the lenses can't handle it.

Almost every lens can benefit, some more than others. My $60 kitlens resolves more with the A77 24MP sensor than with the 12 MP sensor found in my A500. In FF terms that's like going from about 28 to 56 MP. We haven't come close yet.

Even the charts all bear it out in the article. The APC sensors are noisier than the FF and you will find that the FF tend to have larger pixels if it didn't matter the measure would be the same for a 25 MP A99 and A77.. its not.

See above, sensor size. If pixel size was that important, the 36MP FF sensor wouldn't keep up with the 24MP sensor from the same manufacturer, released around the same time.

If pixel size was important, the RX10/RX100II/RX100 sensors wouldn't be as efficient per unit area as the best FF sensors out there, with pixels up to 9 times smaller. In the end, as also shown by DXO measurements, there is no real correlation between pixel size and shot noise per unit sensor area in practise. An extreme case from the sensor makers themselves:

Nordstjernen
Nordstjernen Veteran Member • Posts: 6,876
Re: Hard to bust the large vs small pixel myth
4

K E Hoffman wrote:

Yes you can't assume a higher res sensor will have more noise than another lower res camera or the previous model.. but it would have more than a sensor with the same design and larger photo reception sites.

The core of misconception:

K. E. Hoffman writes: " ... but it would have more than a sensor with the same design and larger photo reception sites."

No, the size/the numbers of photo reception sites does NOT affect the amount of noise for sensors of the same size! What matters is the light-gathering power of the sensor, that is the surface area, which is given by the physical size of the sensor.

So for a 12 Mp, 24 Mp and 36 Mp FF sensor the amount of digital noise will be more or less the same, but the characteristics of the noise might be a bit different.

The same would be true for a 12 Mp, 24 Mp and 36 Mp aps-c sensor.

You can't compare pixel by pixel. You have to compare pictures of the same size. Then the noise level will be almost identical, but the high res sensors will show more detail.

The main reason for this is better micro lenses with almost no gap between the photo sites and also much better noise reduction at the A/D level, before the raw files are written.

 Nordstjernen's gear list:Nordstjernen's gear list
Sony Alpha a99 Sony Alpha 7
Nordstjernen
Nordstjernen Veteran Member • Posts: 6,876
The only problem is that ... not correct
2

Slideshow Bob wrote:

… with the highest resolution sensor again producing the most noise, and the lowest resolution one producing the least.

If you compare pixel to pixel smaller photosites will look noisier, but there are many more of them, gathering useable image data!

So you have to look at the total noise level from the sensor as a unit! The total amount of noise from the sensor will be more or less the same, no matter megapixel figures!

 Nordstjernen's gear list:Nordstjernen's gear list
Sony Alpha a99 Sony Alpha 7
TrojMacReady
TrojMacReady Veteran Member • Posts: 8,696
Re: The only problem is that...
5

Slideshow Bob wrote:

…it actually doesn't appear to work like that. Take Nikon's current FX lineup as an example, and you'll find that when listed in order of noise levels (best to worst) at the image level (so per area rather than per pixel), you'd get this...

Df

D4

D610

D800

…which is actually proving that for a given sensor size, smaller pixels are actually more noisy. You can't resize a D800 image to DF size and achieve the same (or less) noise without applying NR. NR is a post processing step, whether it's done in camera or on a PC.

No, you're comparing different sensor technologies (Sony and Nikon/Renesas). DXO measurements show that at a given exposure and display size (no NR), midtones and highlights (shot noise) score about equal and a benefit in shadows due to read noise. Which has everything to do with sensor technology.

The D800E and D600 score about the same in shadows, midtones and highlights. Same sensor maker, released around the same time, 50% larger pixels in the latter.

Listing Canon FF cameras in the same fashion produces this...

1Dx

6D

5DIII

… with the highest resolution sensor again producing the most noise, and the lowest resolution one producing the least.

The resolution differences between those 3 cameras are trivial when you realize that there's cameras with pixel differences larger than a factor 9, scoring similar in quantum effieciency. Any conclusion based on that example, can be and likely is influenced by 1 or more of the following factors:

  • margin of measuring error (1/3 stop for DXOmark figures)
  • surrounding electronics used (price difference)
  • sensor technology used (release date)

And midtones and highlight measurements score within a third of a stop, only the shadows really differ.

K E Hoffman
K E Hoffman Senior Member • Posts: 4,841
Re: Hard to bust the large vs small pixel myth
2

Nordstjernen wrote:

K E Hoffman wrote:

Yes you can't assume a higher res sensor will have more noise than another lower res camera or the previous model.. but it would have more than a sensor with the same design and larger photo reception sites.

The core of misconception:

K. E. Hoffman writes: " ... but it would have more than a sensor with the same design and larger photo reception sites."

No, the size/the numbers of photo reception sites does NOT affect the amount of noise for sensors of the same size! What matters is the light-gathering power of the sensor, that is the surface area, which is given by the physical size of the sensor.

So for a 12 Mp, 24 Mp and 36 Mp FF sensor the amount of digital noise will be more or less the same, but the characteristics of the noise might be a bit different.

The same would be true for a 12 Mp, 24 Mp and 36 Mp aps-c sensor.

You can't compare pixel by pixel. You have to compare pictures of the same size. Then the noise level will be almost identical, but the high res sensors will show more detail.

The main reason for this is better micro lenses with almost no gap between the photo sites and also much better noise reduction at the A/D level, before the raw files are written.

Yes the optimization of the light gathering surface is a key game changer.  Also the amount of sensor area below the lenses helps. General improvement in sensor sites so they function with less noise. Putting A/D converters closer to the sensor etc.  They all counteract the SNR issues with smaller and smaller light collection units.

That's why you can't compare sensors of different design and assume the pixel size on the Sensor will cause a predictable increase in noise  But same design with larger pixels means better SNR ...

Each sensor site has some natural noise in it.  If someone wants to show me that a larger sensor site has more noise that might be an equalizer.. BUT there are other places in the system that add noise.  The support circuits for each site.. circuit paths before the A/D converters.   Its like fixed costs in a business.. its there.  So when the light collecting ability goes down.. but the base system noise does not.. the SNR gets worse and it shows up when there is less light to collect.. thus we try to find a place to over expose but not blow out..

Noise does not happen at the SENSOR level.. it happens and the Sensor Site..

So if I take the same sensor area and double the resolution.. I cut the light going to each sensor site.. If I do not add in an improvement to lower noise or optimize light collected we get pixel more likely to have a reading that is not accurate.  At some point I might get to 100% efficiency all pixels hitting the lens get collected.  Then I cut the sensor size down.. I just get less light and I can only look at ways to reduce noise in the system.

-- hide signature --

K.E.H. >> Shooting between raindrops in WA<<

 K E Hoffman's gear list:K E Hoffman's gear list
Canon EOS 450D Nikon 1 J1 Sony Alpha a6300 Sony 100mm F2.8 Macro Sony E 18-200mm F3.5-6.3 OSS +4 more
TrojMacReady
TrojMacReady Veteran Member • Posts: 8,696
Re: Hard to bust the large vs small pixel myth
2

You're talking about read noise in general, which indeed scales to some small extend with resolution (not fixed per pixel). Current sensors are largely shot noise limited. For long exposure night shots, read noise can become dominant, but as said, optimizations counteract that in practise. Sensors like found in the RX10 with just 2.4 micron pixels, already show incredibly low read noise, even compared to sensors found in recent DSLR's. That's the equivalent of about 63 MP and 147 MP on an APS-C and FF sensor respectively.

Which means that again in practise, higher resolution APS-C or FF sensors based on that existing technology, would show no real downside in terms of noise, despite much higher resolutions.

Maxxuman Senior Member • Posts: 1,700
How about "This article shouldn't have to be read by everybody"...
3

except pixel peepers...

-- hide signature --

Barry

Nordstjernen
Nordstjernen Veteran Member • Posts: 6,876
Re: How about "This article shouldn't have to be read by everybody"...

Maxxuman wrote:

except pixel peepers...

The ultimate serious camera owner's activity!

100% view = reality!

 Nordstjernen's gear list:Nordstjernen's gear list
Sony Alpha a99 Sony Alpha 7
Maxxuman Senior Member • Posts: 1,700
Re: How about "This article shouldn't have to be read by everybody"...

Nordstjernen wrote:

Maxxuman wrote:

except pixel peepers...

The ultimate serious camera owner's activity!

That's OK, mine's photography.

-- hide signature --

Barry

Allan Olesen Senior Member • Posts: 3,358
Re: How about "This article shouldn't have to be read by everybody"...

Maxxuman wrote:

except pixel peepers...

I agree. Only pixel peepers have the misconceptions that the article corrects.

...or perhaps that was not what you meant?

tbcass
tbcass OP Forum Pro • Posts: 24,941
Re: Hard to bust the large vs small pixel myth
3

K E Hoffman wrote:

Nordstjernen wrote:

tbcass wrote:

" Resolution doesn’t increase noise

Increasing resolution (MPixels) for a given sensor size has no direct impact on image noise. In fact, some of the lowest noise cameras (Nikon D800, Nikon D3x, Pentax K-5, Nikon D7000, Sony Alpha 580) have relatively high resolutions. See Figure 1a."

I think the myth that large pixels gives a cleaner image for sensors of a given size will live on for years. My question is: Why trade resolution for nothing?

There will still be a lot of posters crying for lower res sensors, thinking that insane high ISO settings that is never used for real life photography would look cleaner!

Physics is still physics.. less light to vs base system noise on smaller pixel more noise is seen.

The writer of the article is a physicist. Do you presume to know more than him.

We are getting to the place where the pixel count is not adding to the image because most of the lenses can't handle it.

Even the charts all bear it out in the article. The APC sensors are noisier than the FF and you will find that the FF tend to have larger pixels if it didn't matter the measure would be the same for a 25 MP A99 and A77.. its not.

If you read the article you will see that the total amount of noise depends on the size of the sensor not the number of pixels. A sensor with large but fewer pixels produces the same amount of noise as a same size sensor with more pixels. It's when the image is scaled to the larger size that the noise is magnified because the image is magnified.

What the article almost obfuscates (it mentions it. then dismisses it) is that the fact that the rule is "broken" so often has to do with all the the other variables.. New Sensor Technology and camera design.. etc that allows an A77 to have a better noise rating that the A700 did. Would not happen if it were the same generation sensor but double the pixel count.

And yes Printing / scaling creates system that averages out noise in most cases..

But a 16 MP A77 with the same sensor technology would have less noise.. I think we have seen that in the sony line.

Absolutely not true. A 24mp image scaled to 16mp will have the same noise as a 16mp image as long as the sensor size and design stays the same.

If you print to the same size less than 11x14 etc does it tend to smooth out.. yes. as you make larger prints or CROP the noise will not be smoothed out.

I also enjoyed the chart where is said 20 MP cell phone camera couldn't happen - Tell that to Nokia with the Lumia 1020 at 41 MP. NEVER SAY NEVER related to technology.. odds are you will get burned.

Yes you can't assume a higher res sensor will have more noise than another lower res camera or the previous model.. but it would have more than a sensor with the same design and larger photo reception sites.

Wrong. I guess you are one of those who will go on believing the myth no matter the facts presented.

-- hide signature --

Tom
Look at the picture, not the pixels
------------
Miss use of the ability to do 100% pixel peeping is the bane of digital photography because it causes people to fret over inconsequential issues.
http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/

 tbcass's gear list:tbcass's gear list
Sony RX100 Sony SLT-A77 Tamron SP AF 90mm F/2.8 Di Macro Sony DT 35mm F1.8 SAM Tamron SP 70-300mm F/4-5.6 Di USD +8 more
Allan Olesen Senior Member • Posts: 3,358
Re: Hard to bust the large vs small pixel myth
2

K E Hoffman wrote:

If you print to the same size less than 11x14 etc does it tend to smooth out.. yes. as you make larger prints or CROP the noise will not be smoothed out.

It is not only downsampling which levels the playing field between low and high MP sensors. The principle is exactly the same in upsampling/enlarging:

If you have the same amount of per pixel noise in two photos with different pixel count, and you enlarge both photos to the same size, the photo with the fewest pixels will have more visible noise.

So you can choose to see the lower per pixel noise of a low MP sensor as an advantage. But in reality, the lower per pixel noise is necessary to outweigh the disadvantage which it has when you scale the photo to its final size.

Keyboard shortcuts:
FForum MMy threads