Reading mode: Light Dark

Updated product pages: Why not pixel density ?

Go to last post
Of course not.
Roland Karlsson said:
tom sugnet said:
Yeah, pixel density means nothing.

All new point&shoots have fabulous picture quality, super crisp, and awesome ISO3200.
Click to expand...
You are totally 100% missing the point.

The reason small sensor cameras has problems with image quality is that the sensors are small. If you had an APS-C camera with the same pixel densioty you would have a fantastic IQ.

--
Roland

support http://www.openraw.org/
(Sleeping - so the need to support it is even higher)

X3F tools : http://www.proxel.se/x3f.html
Click to expand...
--
Feel free to visit my photo sites:
http://tom.st , http://www.foto.tom.st

 
tom sugnet said:
Of course not.
Click to expand...
OK - so why dont they do the small sensor camera with a smaller pixel density then?

Because the number of pixels is important.

For a given sensor size, if you start at a small number of pixels ... then the image quality will go up, up, up, ... when increasing the number of pixels. Until you hit an optimal number of pixels, then it will go down again.

What that optimum is depends on sensor size, ISO number, technology and your subjective opinion.

--
Roland

support http://www.openraw.org/
(Sleeping - so the need to support it is even higher)

X3F tools : http://www.proxel.se/x3f.html
 
ljfinger said:
tom sugnet said:
Of course not.
Click to expand...
You are wrong, and I can and have proven it with test results.

--
Lee Jay
(see profile for equipment)
Click to expand...
Yes .. and it has been done several times.

You can easily simulate a FF camera with 500 MP using a small sensor camera and stitching.

I promise you that the 500 MP stitched image blows away the 25 MP FF camera any day and printed at any size.

The myth that few pixels are better than many is simply false.

OK - you can construct situations where it is true. But an FF sensor with the pixel density of a small sensor camera is not one of those. At least not for reasonable ISO values.

Few pixels are good for
  1. high speed photography
  2. extremely high ISO photography
  3. saving storage
  4. very small sensors
--
Roland

support http://www.openraw.org/
(Sleeping - so the need to support it is even higher)

X3F tools : http://www.proxel.se/x3f.html
 
Roland Karlsson said:
ljfinger said:
tom sugnet said:
Of course not.
Click to expand...
You are wrong, and I can and have proven it with test results.

--
Lee Jay
(see profile for equipment)
Click to expand...
Yes .. and it has been done several times.

You can easily simulate a FF camera with 500 MP using a small sensor camera and stitching.

I promise you that the 500 MP stitched image blows away the 25 MP FF camera any day and printed at any size.

The myth that few pixels are better than many is simply false.

OK - you can construct situations where it is true. But an FF sensor with the pixel density of a small sensor camera is not one of those. At least not for reasonable ISO values.

Few pixels are good for
  1. high speed photography
  2. extremely high ISO photography
  3. saving storage
  4. very small sensors
Click to expand...
A FF sensor that has the same tech and pixel density as the sensor in the Canon G12, would kill any current existing sensor in terms of the IQ of the photo. There would be nothing even remotely close.
 
Roland Karlsson said:
You can easily simulate a FF camera with 500 MP using a small sensor camera and stitching.

I promise you that the 500 MP stitched image blows away the 25 MP FF camera any day and printed at any size.
Click to expand...
Yes, but the stitched 500mp image will also have much, much higher resolution than an image from a 500mp FF camera using real world lenses.
 
Great Bustard said:
A FF sensor that has the same tech and pixel density as the sensor in the Canon G12, would kill any current existing sensor in terms of the IQ of the photo. There would be nothing even remotely close.
Click to expand...
Yes, the resolution would be nice of course (with good lenses..), but the G12's 'flat' read noise at about 4.4e- isn't really that low. If we downsample/normalize a FF image with G12 pixels to e.g. 10mp, then the normalized read noise (that scales with the linear pixel size) will be about 20e-, which is app. the same as a D3s iso100 image normalized to 10mp, but at higher ISOs the D3s images will have much lower read noise than than the normalized images from the FF camera with G12 pixels.
 
Steen Bay said:
Great Bustard said:
A FF sensor that has the same tech and pixel density as the sensor in the Canon G12, would kill any current existing sensor in terms of the IQ of the photo. There would be nothing even remotely close.
Click to expand...
Yes, the resolution would be nice of course (with good lenses..), but the G12's 'flat' read noise at about 4.4e- isn't really that low. If we downsample/normalize a FF image with G12 pixels to e.g. 10mp, then the normalized read noise (that scales with the linear pixel size) will be about 20e-, which is app. the same as a D3s iso100 image normalized to 10mp, but at higher ISOs the D3s will have much lower read noise than than the normalized image from the FF camera with G12 pixels.
Click to expand...
The read noise is very low for a CCD, but it doesn't intrinsically scale with pixel size the same way as it would for a MOS sensor. The read noise is largely controlled by teh size of the source follower charge amplifier transistor, which is in the pixel in a mos sensor, but a separate unit at the end of the charge coupled shift register on a CCD. I suspect that the G11/G12 sensor is the limit of what can be done with current CCD fab lines, in terms of shrinking that transistor, and that there will be no further CCD lines being built, all the action is in MOS.

So, I'm not sure it makes much sense to use the G12 as an example of what could be done, since it's likely that a MOS sensor with that pixel geometry would have much lower read noise. Whether it would, with current technology, have the same QE is an open question. When DxO gives data on some of the new P&S MOS sensors, I'll give it the sensorgen treatment.
--
Bob
 
bobn2 said:
Steen Bay said:
Great Bustard said:
A FF sensor that has the same tech and pixel density as the sensor in the Canon G12, would kill any current existing sensor in terms of the IQ of the photo. There would be nothing even remotely close.
Click to expand...
Yes, the resolution would be nice of course (with good lenses..), but the G12's 'flat' read noise at about 4.4e- isn't really that low. If we downsample/normalize a FF image with G12 pixels to e.g. 10mp, then the normalized read noise (that scales with the linear pixel size) will be about 20e-, which is app. the same as a D3s iso100 image normalized to 10mp, but at higher ISOs the D3s will have much lower read noise than than the normalized image from the FF camera with G12 pixels.
Click to expand...
The read noise is very low for a CCD, but it doesn't intrinsically scale with pixel size the same way as it would for a MOS sensor. The read noise is largely controlled by teh size of the source follower charge amplifier transistor, which is in the pixel in a mos sensor, but a separate unit at the end of the charge coupled shift register on a CCD. I suspect that the G11/G12 sensor is the limit of what can be done with current CCD fab lines, in terms of shrinking that transistor, and that there will be no further CCD lines being built, all the action is in MOS.

So, I'm not sure it makes much sense to use the G12 as an example of what could be done, since it's likely that a MOS sensor with that pixel geometry would have much lower read noise. Whether it would, with current technology, have the same QE is an open question. When DxO gives data on some of the new P&S MOS sensors, I'll give it the sensorgen treatment.
Click to expand...
Yes, it'll be interesting to see some data from the new small BSI CMOS sensors. Don't know what the data will show, but it's my impression that people generally aren't too impressed with the IQ.
 
Steen Bay said:
bobn2 said:
Steen Bay said:
Great Bustard said:
A FF sensor that has the same tech and pixel density as the sensor in the Canon G12, would kill any current existing sensor in terms of the IQ of the photo. There would be nothing even remotely close.
Click to expand...
Yes, the resolution would be nice of course (with good lenses..), but the G12's 'flat' read noise at about 4.4e- isn't really that low. If we downsample/normalize a FF image with G12 pixels to e.g. 10mp, then the normalized read noise (that scales with the linear pixel size) will be about 20e-, which is app. the same as a D3s iso100 image normalized to 10mp, but at higher ISOs the D3s will have much lower read noise than than the normalized image from the FF camera with G12 pixels.
Click to expand...
The read noise is very low for a CCD, but it doesn't intrinsically scale with pixel size the same way as it would for a MOS sensor. The read noise is largely controlled by teh size of the source follower charge amplifier transistor, which is in the pixel in a mos sensor, but a separate unit at the end of the charge coupled shift register on a CCD. I suspect that the G11/G12 sensor is the limit of what can be done with current CCD fab lines, in terms of shrinking that transistor, and that there will be no further CCD lines being built, all the action is in MOS.

So, I'm not sure it makes much sense to use the G12 as an example of what could be done, since it's likely that a MOS sensor with that pixel geometry would have much lower read noise. Whether it would, with current technology, have the same QE is an open question. When DxO gives data on some of the new P&S MOS sensors, I'll give it the sensorgen treatment.
Click to expand...
Yes, it'll be interesting to see some data from the new small BSI CMOS sensors. Don't know what the data will show, but it's my impression that people generally aren't too impressed with the IQ.
Click to expand...
I never take much credence from 'people' unless those 'people' are identified and I know what they are saying. There was a case a bit ago where someone posted a clip form an article to suggest that the image quality of the new 16MP compacts was poor. I took the trouble to find the complete article, and what the reviewers had done was post samples from the camera and its predecessor and ask the readers to make a blind preference (they didn't know which image came from which camera) - the preference was overwhelmingly for the 16MP camera.

That is part of this whole debate, many people diss high pixel density cameras because they think it is the thing to do, and they think its the thing to do because of the hangovers of Phil Askey's misguided and mendacious campaign, of which the pixel density metric in DPR's data was a part.
--
Bob
 
Roland Karlsson said:
Hmmmmm ... this is my take on it.

You had an example where pixel density is interesting. That is if you have cameras of different sensor sizes (like APS-C and FF) and can use the same lens (or the same focal length lens) and take some image (probably wild life) where you crop a lot.

Right?

Then I wonder - why in the first place are you interested in using the larger sensor camera for this application in the first place?
Click to expand...
In the first place the larger sensor camera was the only one I had ;-)
 
Steen Bay said:
Great Bustard wrote:

Yes, the resolution would be nice of course (with good lenses..), but the G12's 'flat' read noise at about 4.4e- isn't really that low.
Click to expand...
Do you often take pictures where this level of read noise matters? I am surprised to see this statement, and I would consider this level to be low.
Steen Bay said:
If we downsample/normalize a FF image with G12 pixels to e.g. 10mp, then the normalized read noise (that scales with the linear pixel size) will be about 20e-,
Click to expand...
I don't understand what normalized read noise is, or what you mean by scaling with linear pixel size. Obviously read noise in sensors does not scale with pixel size and if anything the read noise tends to drop with pixel size. Maybe you can explain better what you mean, if you don't mind.
 
Steen Bay said:
Yes, it'll be interesting to see some data from the new small BSI CMOS sensors. Don't know what the data will show, but it's my impression that people generally aren't too impressed with the IQ.
Click to expand...
BSI is expected to improve image quality in sensors with pixel pitch of 1100 nm and below. For larger pixels the cost benefit ratio deteriorates, but can still be beneficial in more subtle ways. FSI pixels are pretty good already at DSLR pixel pitches. BSI can also allow better pixel functionality, such as higher dynamic range, or global shutter (aka mirrorless operation) since added circuitry in the pixel has less impact on pixel fill factor.
 
Eric Fossum said:
Steen Bay said:
Great Bustard wrote:

Yes, the resolution would be nice of course (with good lenses..), but the G12's 'flat' read noise at about 4.4e- isn't really that low.
Click to expand...
Do you often take pictures where this level of read noise matters? I am surprised to see this statement, and I would consider this level to be low.
Click to expand...
Yes, the G12 has a very fine sensor with a rather impressive QE, but nevertheless its read noise is relatively high if compared to e.g. the APS-C sensor used in K-5 and D7000, that has an even lower read noise and at the same time a much higher saturation capacity (and therefore higher DR).
Eric Fossum said:
Eric Fossum said:
If we downsample/normalize a FF image with G12 pixels to e.g. 10mp, then the normalized read noise (that scales with the linear pixel size) will be about 20e-,
Click to expand...
I don't understand what normalized read noise is, or what you mean by scaling with linear pixel size. Obviously read noise in sensors does not scale with pixel size and if anything the read noise tends to drop with pixel size. Maybe you can explain better what you mean, if you don't mind.
Click to expand...
What I meant was that if we downsample or 'normalize' (like in DxO's 'print' figures) the image from a hypothetical FF sensor with G12 sized pixels to e.g. 10mp (in order to compare at the same scale), then the app. 4.4e- read noise that the G12 has is 'equivalent' to about 20e- on a native 10mp FF sensor, which isn't really that low (and especially not at high ISOs). I think that my rather crude calculation is app. correct, but would appreciate if Bob had a comment here.
 
Steen Bay said:
Eric Fossum said:
Steen Bay said:
Great Bustard wrote:

Yes, the resolution would be nice of course (with good lenses..), but the G12's 'flat' read noise at about 4.4e- isn't really that low.
Click to expand...
Do you often take pictures where this level of read noise matters? I am surprised to see this statement, and I would consider this level to be low.
Click to expand...
Yes, the G12 has a very fine sensor with a rather impressive QE, but nevertheless its read noise is relatively high if compared to e.g. the APS-C sensor used in K-5 and D7000, that has an even lower read noise and at the same time a much higher saturation capacity (and therefore higher DR).
Click to expand...
I think this is a commentary on the recent progress made by the Japanese sensor manufacturers. 4 e- is certainly low for a CCD (they tend to average 10 and up) and would have been low by any standards three years ago - it is the same level as, for instance the 1DIII. Since then Sony, Canon and Nikon have made sensors with 3-2, and maybe sub-2 electrons - mostly, I suspect (although Eric is in a better place to give an authoritative opinion) as a result of geometry shrinks. So, per pixel, for a CCD, the read noise is low.

I think what Eric is getting at though, is that there is a case of diminishing returns with lowering of read noise in terms of practical photography. If you take a photo using 9 stops of DR from a noise floor of 1 e- per pixel, then the highlights will have 512 electrons, and an SNR of 22. The results are unlikely to be usable.
Steen Bay said:
Steen Bay said:
Steen Bay said:
If we downsample/normalize a FF image with G12 pixels to e.g. 10mp, then the normalized read noise (that scales with the linear pixel size) will be about 20e-,
Click to expand...
I don't understand what normalized read noise is, or what you mean by scaling with linear pixel size. Obviously read noise in sensors does not scale with pixel size and if anything the read noise tends to drop with pixel size. Maybe you can explain better what you mean, if you don't mind.
Click to expand...
What I meant was that if we downsample or 'normalize' (like in DxO's 'print' figures) the image from a hypothetical FF sensor with G12 sized pixels to e.g. 10mp (in order to compare at the same scale), then the app. 4.4e- read noise that the G12 has is 'equivalent' to about 20e- on a native 10mp FF sensor, which isn't really that low (and especially not at high ISOs). I think that my rather crude calculation is app. correct, but would appreciate if Bob had a comment here.
Click to expand...
My view would be that a 10MP sensor is a 10MP sensor and read noise is directly comparable between images of the same pixel count (just as DxO does, the 'normalise' to a pixel count -8MP - not a sensor size). The FF sensor with G12 size pixels would, according to my rough calculations, have about 200MP, which, if it had the same read noise of 4.4 e- per pixel would result in about 20 e- per 10MP output pixel after downsampling - which might be what you mean (still not too bad, the Leica M9 has a read noise of about 15 e- per pixel, and would downsample to about 20 at 10MP also. On the other hand, the M9 is collecting about 56000 electrons over 1 ten-millionth of the image, whereas 20 G12 pixels would collect about 200000, so the G12 FF sensor would end up with about two stops more DR than the M9. So far as CCD tech goes, it seems the smaller pixels are winning.

Of course, as I said, this is CCD. If it were MOS read noise intrinsically scales down with the area, due to the reduced capacitance of the read transistor gate and floating diffusion - we should remember though that these are designed things and smaller pixels are not produced simply by scaling designs for big ones. The state of the art at the moment (with respect to read noise) seems to be the Canon 1DIV, with 1.5 e- in a 5.7μm pixel. For the 2.03μm pixels of the G12 - if they were simply scaled 1DIV pixels and somehow the designers and manufacturers managed somehow to scale the read noise with area, then we might expect about 0.2 e- per pixel, which when downsampled to10MP from a 200MP FF sensor full of them would be about 0.9 e-. I don't expect the 2μm MOS pixels, when we see some figures, to be close to that good.
--
Bob
 
bobn2 said:
I think what Eric is getting at though, is that there is a case of diminishing returns with lowering of read noise in terms of practical photography.
Click to expand...
There are special cases where read noise can be of critical importance:

http://mallincam.tripod.com/

--
Lee Jay
(see profile for equipment)
 
bobn2 said:
Steen Bay said:
Eric Fossum said:
Steen Bay said:
Great Bustard wrote:

Yes, the resolution would be nice of course (with good lenses..), but the G12's 'flat' read noise at about 4.4e- isn't really that low.
Click to expand...
Do you often take pictures where this level of read noise matters? I am surprised to see this statement, and I would consider this level to be low.
Click to expand...
Yes, the G12 has a very fine sensor with a rather impressive QE, but nevertheless its read noise is relatively high if compared to e.g. the APS-C sensor used in K-5 and D7000, that has an even lower read noise and at the same time a much higher saturation capacity (and therefore higher DR).
Click to expand...
I think this is a commentary on the recent progress made by the Japanese sensor manufacturers. 4 e- is certainly low for a CCD (they tend to average 10 and up) and would have been low by any standards three years ago - it is the same level as, for instance the 1DIII. Since then Sony, Canon and Nikon have made sensors with 3-2, and maybe sub-2 electrons - mostly, I suspect (although Eric is in a better place to give an authoritative opinion) as a result of geometry shrinks. So, per pixel, for a CCD, the read noise is low.

I think what Eric is getting at though, is that there is a case of diminishing returns with lowering of read noise in terms of practical photography. If you take a photo using 9 stops of DR from a noise floor of 1 e- per pixel, then the highlights will have 512 electrons, and an SNR of 22. The results are unlikely to be usable.
bobn2 said:
bobn2 said:
bobn2 said:
If we downsample/normalize a FF image with G12 pixels to e.g. 10mp, then the normalized read noise (that scales with the linear pixel size) will be about 20e-,
Click to expand...
I don't understand what normalized read noise is, or what you mean by scaling with linear pixel size. Obviously read noise in sensors does not scale with pixel size and if anything the read noise tends to drop with pixel size. Maybe you can explain better what you mean, if you don't mind.
Click to expand...
What I meant was that if we downsample or 'normalize' (like in DxO's 'print' figures) the image from a hypothetical FF sensor with G12 sized pixels to e.g. 10mp (in order to compare at the same scale), then the app. 4.4e- read noise that the G12 has is 'equivalent' to about 20e- on a native 10mp FF sensor, which isn't really that low (and especially not at high ISOs). I think that my rather crude calculation is app. correct, but would appreciate if Bob had a comment here.
Click to expand...
My view would be that a 10MP sensor is a 10MP sensor and read noise is directly comparable between images of the same pixel count (just as DxO does, the 'normalise' to a pixel count -8MP - not a sensor size). The FF sensor with G12 size pixels would, according to my rough calculations, have about 200MP, which, if it had the same read noise of 4.4 e- per pixel would result in about 20 e- per 10MP output pixel after downsampling - which might be what you mean (still not too bad, the Leica M9 has a read noise of about 15 e- per pixel, and would downsample to about 20 at 10MP also. On the other hand, the M9 is collecting about 56000 electrons over 1 ten-millionth of the image, whereas 20 G12 pixels would collect about 200000, so the G12 FF sensor would end up with about two stops more DR than the M9. So far as CCD tech goes, it seems the smaller pixels are winning.

Of course, as I said, this is CCD. If it were MOS read noise intrinsically scales down with the area, due to the reduced capacitance of the read transistor gate and floating diffusion - we should remember though that these are designed things and smaller pixels are not produced simply by scaling designs for big ones. The state of the art at the moment (with respect to read noise) seems to be the Canon 1DIV, with 1.5 e- in a 5.7μm pixel. For the 2.03μm pixels of the G12 - if they were simply scaled 1DIV pixels and somehow the designers and manufacturers managed somehow to scale the read noise with area, then we might expect about 0.2 e- per pixel, which when downsampled to10MP from a 200MP FF sensor full of them would be about 0.9 e-. I don't expect the 2μm MOS pixels, when we see some figures, to be close to that good.
Click to expand...
Agree with all that, and yes, that was what I meant when talking about the 20e- 'normalized' read noise, but the way I put it was probably a bit confusing. Thanks for the clarification. ;-)
 
ljfinger said:
bobn2 said:
I think what Eric is getting at though, is that there is a case of diminishing returns with lowering of read noise in terms of practical photography.
Click to expand...
There are special cases where read noise can be of critical importance:
Click to expand...
Indeed, special cases. Astronomical requirements are a bad guide to what makes a good sensor for general photography. It's at the root of all this 'per pixel' thinking - which is valid for astronomy.
Lee Jay said:
http://mallincam.tripod.com/
Click to expand...
http://www.ast.cam.ac.uk/research/instrumentation.surveys.and.projects/lucky.imaging/lucky.imaging.methods

http://www.ast.cam.ac.uk/research/instrumentation.surveys.and.projects/lucky.imaging/l3ccd.technology
--
Bob
 
Le Kilt said:
In the first place the larger sensor camera was the only one I had ;-)
Click to expand...
Would you accept pixel pitch instead?

Personally I prefer easy to understand concrete figures.

Pixel density feels so abstract. I cannot really relate to it IMHO.

But pixel pitch, number of pixels and sensor size are three measures I easily understand.

--
Roland

support http://www.openraw.org/
(Sleeping - so the need to support it is even higher)

X3F tools : http://www.proxel.se/x3f.html
 
You must log in or register to reply here.

Similar threads

JimKasson
Noise, Dynamic Range, Cropping, and Print Size
2
Replies
24
Views
2K
JimKasson
JimKasson
I
Why Sensors Read Slow and the Sony IMX366 44MP 26 fps Sensor
2
Replies
27
Views
3K
Interceptor121
I
JimKasson
How much bigger is 33x44mm MF than FF?
6 7 8
Replies
149
Views
989
Doppler9000
D
JimKasson
Why we own cameras, and why it matters
2
Replies
38
Views
3K
DavidMillier
DavidMillier
U
X3F prematurely sentenced to ...
Replies
4
Views
5
Ulverus
U

Keyboard shortcuts

f
Forum
m
My threads
Mobile site
About
Editorial content
Cameras & Lenses
Community
All content, design, and layout are Copyright © 1998–2025 Digital Photography Review All Rights Reserved.
Reproduction in whole or part in any form or medium without specific written permission is prohibited.
When you use DPReview links to buy products, the site may earn a commission.
©GPS Media - Guides, Products, Services.
Back
Top