Relative Pixel Density - don't expect miracles from 40D

[Andrew dB]

The problem is that you are doing a comparison while changing two
different variables at once. You can't be sure whether the superior
image quality of the larger sensors is down to pixel size or sensor
size because both have been changed. Any science teacher will tell
you that makes it almost impossible to draw a worthwhile conclusion.

The speculation in the rest of your post is interesting, but I take
issue with the statement above. I deliberately calculated an
entirely comparable figure for all the sensors precisely to get away
from the confusion of different sensor sizes.

True, but then the image quality comparison is done by looking at the output from the entire sensor which complicates things because there is no way to tell whether differences are caused by the different pixel sizes or the overall sensor size.

If you look again you will see that I worked out the number of pixels
per unit of sensor area - i.e MP per square inch. This figure is
effectively an analogue for pixel size, but can be directly compared
without reference to actual sensor sizes.

Of course if you used the low 9.6 MP/sq in pixel density of the FF 5D
on a 1.6 sensor you would get a rather low resolution image, but of
course that is why you need bigger sensors if you want to get the
benefits of using bigger pixels, while retaining reasonably high
resolution.

Fred

I agree with the sentiment that bigger sensors (and lenses) will give better results but the origin of this descussion was the claim that small pixels were a bad thing that should be avoided on DSLRs. By comparing equal sensor areas in terms of both pixel density and image quality, we can assess whether a hypothetical 5D successor that was based on the 400D sensor (with similar size pixels) would be a good thing in terms of overall image quality without the comparison being complicated by the issue of sensor size.

 

[John Sheehy]

so my original point still stands, that for a given resolution the
larger sized sensor, i.e. the one with lower pixel density, will
perform better with respect to total image quality.

Was that your original point? I thought your original point was that high pixel densities are bad, which is another thing entirely.

--
John

 

[carlk]

Your comparison is meanless and the conclusion is just plainly wrong. Do you imply if Canon make 2MP 40D it will have best IQ of them all? Numbers don't lie but people often do.

Just out of interest, and further to my recent thread suggesting the
40D would be a better camera if kept to 8MP, I have worked out the
relative pixel density of several models representative of different
sensor sizes and resolutions. I have used Canons official specs for
sensor dimensions.

(I'm sure something similar has been posted before, but maybe not
specifically in the context of expected 40D performance, so I thought
it was worth posting.)

I have expressed these figures in MegaPixels per square inch, as area
is the key measure for the light gathering ability of a sensor/pixel.

What I get is as follows:

5D ------------- 9.60 MP per sq in (12.8MP sensor, 1.0 crop factor)

1D MK3 ------ 12.27 MP per sq in (10MP sensor, 1.3 crop factor)

1Ds MK2 ----- 12.53 MP per sq in (16.7MP sensor, 1.0 crop factor )

30D ----------- 15.59 MP per sq in (8.2MP sensor, 1.6 crop factor)

400D ---------- 19.97 MP per sq in (10MP sensor, 1.6 crop factor)

I think it is striking how closely this measure seems to corresponds
with the reported relative IQ (Image Quality) from these cameras.
This number also gives a good measure of how high a demand the
camera makes on lens resolution, with even the best lenses possibly
being asked for too much by very high pixel densities.

In particular look at the 5D - is it any wonder that most people
think this has the best IQ of any current DSLR, and gets very good
results from a wide range of lenses.

In interpreting these figures you do have to make some allowance for
different sensor generations. For example the more recent 400D is
generally accepted as being close to the 30D in IQ (though IMO not as
good) despite the higher resolution, but does have a much more recent
sensor design.

However all things being equal, if we assume the 40D will be 10MP and
will have all the sensor improvements seen in the 1D MK3 (by no means
certain) we are looking at relative pixel densities, and therefore
relative pixel sizes in a ratio of 12.27 versus 19.97.

This is a ratio of 1.62 in favour of the 1D MK 3 even at equal
resolution and if all the same sensor improvements are made. (For an
8MP 40D the ratio would have been a much closer 1.27)

I think this should put in perspective the pipedreams of those who
are expecting DR and low light performance for a 10MP 40D close to
that of the 1D MK3. I can't see any way this can happen. I just
hope the 40D doesn't end up as a 12MP 1.6 sensor.

Sure there can be more agressive noise removal, but this will impact
on ultimate IQ, which I believe is what happens with the 400D
compared to the 20D/30D.

Fred

PS - the earlier thread is here:

http://forums.dpreview.com/forums/read.asp?forum=1019&message=24194540

 

[DSPographer]

I agree that there is still opportunity for improvements in category A below. I have read about new video camera sensors that have one A/D for each sensor row; this type of architecture has the potential to reduce A/D noise substantially greatly increasing the D/R of low ISO images.

Some of us are old enough and with good enough memories to remember
the 10D to 20D transisition.

Even though the 10D had bigger cells, the 20D had lower noise.

Oh, my gosh, how did they do that?

A) the 10D used 2 channels of A/Ds operating at 24 MHz
....the 20D uses 4 channels of A/Ds operating at 16 MHz
....Most of the noise improvement comes from the reconfiguration of A/Ds

B) a host of imporvements in sensor technology
...1) better microlenses
...2) different metalization
...3) smaller transistor footprint compared to cell area
...4) different color microfilters

There is nothing preventing Canon from doing more A/Ds, and indeed if
higher speeds (7 fps) or more bits/read (12-> 14) then slower A/Ds
will be "de rigor".

Also, there is nothing preventing even better miroclensing, and
better transistor overheads, or better wiring overheads.

Finally, there is nothing preventing Canon from doing something in
the Noise Abatement catagory either.
--
Mitch

 

[upnorthfar]

You managed to create demand here.

High ISO is not relevant to my landscape stuff. Yes I want a 35DL 40Mp ISO25-200 Landscape camera!

"The lower pixel noise of a bigger pixel, displayed bigger, is wrong
over a larger area, and it is also wrong in that it doesn't reflect
the detail beneath it like more smaller pixels would; another kind of
noise that generally isn't talked about."

I get this. But why we do not see high end solutions based on this
thinking, say a full frame or 1.5 sensor with a small digicam pixel
density?

High expense and low demand. I'm trying to build the demand.

--
John

 

[Fred Briggs]

Your comparison is meanless and the conclusion is just plainly wrong.
Do you imply if Canon make 2MP 40D it will have best IQ of them all?

Why is that so many people on this thread seem unable to grasp that low pixel density is not necessarily the same as low resolution. The 5D has the lowest pixel density of the cameras I listed , yet has the second highest resolution - due to the large FF sensor.

I am simply observing that the sensors with the lowset pixel density seem to have the highest image quality. However, I've never suggested or even implied that you should achieve this by putting ridiculously low numbers of pixels on a sensor.

Obviously to get the image quality associated with large pixels, while retaining an adequate resolution, you need a bigger sensor in order to fit in enough of these bigger pixels.

Of course Canon could also stop increasing the number of pixels to unnecessarily high levels. That way improvements in sensor design could actually translate into significantly improved DR and noise performance instead of being primarily used up offset the losses due to higher pixel density.

That is why I compared the 1D3 with 10MP on an APS-H sensor with the rumoured 40D with 10MP on the smaller APS-C sensor and said that we should not expect similar DR and high ISO performance from the smaller sensor, even if it incorporated all the same improvements as the 1D3.

Nowhere did I suggest reducing resolution to very low levels would improve image quality. In a previous post I did say that I thought that 8MP was adequate resolution for this class of camera, and forgoing the increase to 10MP could be worthwhile in terms of better image quality. That is not to say that a 10MP 40D won't manage some IQ improvements, just that they could be bigger at 8MP.

Numbers don't lie but people often do.

I hope that the above comment is not accusing me of lying - because if it is then I expect an apology.

Fred

 

[upnorthfar]

You got me thinking this really.
5D sensor area is something like 10 times bigger than G6 digicam with 7,1Mp.

If there was 10 G6 sensors joined side by side to fill the area instead of the 5D 12Mp sensor...
The resolution advantage would be huge!

And noise.. Wouldn't it be actually better than 12Mp 5D in normal 50-200 sensitivities? Per pixel more yes, but there would be SO many pixels and the 12Mp is only a tiny crop of the imaginary 71Mp full size sensor...

And color tones and such.. The huge amount of pixels would perhaps make a lot smoother tonal changes etc..?

"The lower pixel noise of a bigger pixel, displayed bigger, is wrong
over a larger area, and it is also wrong in that it doesn't reflect
the detail beneath it like more smaller pixels would; another kind of
noise that generally isn't talked about."

I get this. But why we do not see high end solutions based on this
thinking, say a full frame or 1.5 sensor with a small digicam pixel
density?

High expense and low demand. I'm trying to build the demand.

--
John

 

[coldbivy]

Your comparison is meanless and the conclusion is just plainly wrong.
Do you imply if Canon make 2MP 40D it will have best IQ of them all?

Why is that so many people on this thread seem unable to grasp that
low pixel density is not necessarily the same as low resolution. The
5D has the lowest pixel density of the cameras I listed , yet has the
second highest resolution - due to the large FF sensor.

I am simply observing that the sensors with the lowset pixel density
seem to have the highest image quality. However, I've never
suggested or even implied that you should achieve this by putting
ridiculously low numbers of pixels on a sensor.

Obviously to get the image quality associated with large pixels,
while retaining an adequate resolution, you need a bigger sensor in
order to fit in enough of these bigger pixels.

Of course Canon could also stop increasing the number of pixels to
unnecessarily high levels. That way improvements in sensor design
could actually translate into significantly improved DR and noise
performance instead of being primarily used up offset the losses due
to higher pixel density.

That is why I compared the 1D3 with 10MP on an APS-H sensor with the
rumoured 40D with 10MP on the smaller APS-C sensor and said that we
should not expect similar DR and high ISO performance from the
smaller sensor, even if it incorporated all the same improvements as
the 1D3.

Nowhere did I suggest reducing resolution to very low levels would
improve image quality. In a previous post I did say that I thought
that 8MP was adequate resolution for this class of camera, and
forgoing the increase to 10MP could be worthwhile in terms of better
image quality. That is not to say that a 10MP 40D won't manage some
IQ improvements, just that they could be bigger at 8MP.

Numbers don't lie but people often do.

I hope that the above comment is not accusing me of lying - because
if it is then I expect an apology.

Fred

I think you've confused any point you were trying to make by asserting that IQ is the same as DR and that high ISO performance is so important as compared to low ISO performance. After all, noise is not an issue at low ISO in any of those cameras and the DR is not all that different.

 

[Fred Briggs]

You got me thinking this really.
5D sensor area is something like 10 times bigger than G6 digicam with
7,1Mp.
If there was 10 G6 sensors joined side by side to fill the area
instead of the 5D 12Mp sensor...
The resolution advantage would be huge!
And noise.. Wouldn't it be actually better than 12Mp 5D in normal
50-200 sensitivities? Per pixel more yes, but there would be SO many
pixels and the 12Mp is only a tiny crop of the imaginary 71Mp full
size sensor...
And color tones and such.. The huge amount of pixels would perhaps
make a lot smoother tonal changes etc..?

This is an interesting suggestion. However I'm not sure that for any normal use the huge additional processing and storage overhead of so many pixels would be worth it, hence I can't see how there would ever be a mass market camera using this approach.

Also, while you may be happy to be effectively limited to 50-200 ISO, I think most people would find this very limiting. i suspect an image composed of 12 x G6 sensors would still look pretty awful at ISO 1600, whereas the large pixel approach of the 1D3 produces highly usable images at ISO 1600 and beyond, while its mere 10MP resolution seems more than adequate for most real world purposes.

I really can't see why so many people think they need higher resolution. An 8MP image can print 17" wide at 200 dpi. Most printing, and viewing on computer screens will be much smaller than this. My 8MP 20D files look great at 1:1 on my 19" computer screen yet at 1:1 I can only see about 1/9th of the picture at a time.

I do accept that resolution is part of the IQ equation, but I tend to take the view that once you have enough resolution, then more is not particularly useful.

I would therefore prefer other components of IQ to be addressed, such as DR and High ISO performance, where there are still deficiencies in all current DSLRs, and even more so in compact cameras.

Fred

 

[upnorthfar]

You got me thinking this really.
5D sensor area is something like 10 times bigger than G6 digicam with
7,1Mp.
If there was 10 G6 sensors joined side by side to fill the area
instead of the 5D 12Mp sensor...
The resolution advantage would be huge!
And noise.. Wouldn't it be actually better than 12Mp 5D in normal
50-200 sensitivities? Per pixel more yes, but there would be SO many
pixels and the 12Mp is only a tiny crop of the imaginary 71Mp full
size sensor...
And color tones and such.. The huge amount of pixels would perhaps
make a lot smoother tonal changes etc..?

This is an interesting suggestion. However I'm not sure that for any
normal use the huge additional processing and storage overhead of so
many pixels would be worth it, hence I can't see how there would ever
be a mass market camera using this approach.

Also, while you may be happy to be effectively limited to 50-200 ISO,
I think most people would find this very limiting. i suspect an
image composed of 12 x G6 sensors would still look pretty awful at
ISO 1600, whereas the large pixel approach of the 1D3 produces highly
usable images at ISO 1600 and beyond, while its mere 10MP resolution
seems more than adequate for most real world purposes.

I really can't see why so many people think they need higher
resolution. An 8MP image can print 17" wide at 200 dpi. Most
printing, and viewing on computer screens will be much smaller than
this. My 8MP 20D files look great at 1:1 on my 19" computer screen
yet at 1:1 I can only see about 1/9th of the picture at a time.

I do accept that resolution is part of the IQ equation, but I tend to
take the view that once you have enough resolution, then more is not
particularly useful.

I would therefore prefer other components of IQ to be addressed, such
as DR and High ISO performance, where there are still deficiencies in
all current DSLRs, and even more so in compact cameras.

Fred

You are right. For one camera solution it's not sensible.

But especially landscape work benefits from extra pixels. And it doesn't require high ISO. Traditionally lot of it has been done with medium or big format cameras.

I wouldn't want this ISO 50-200 camera for my birding. But for the landscapes.. ..YES!

If we decide that full 35mm frame is the size where we want to be maybe the two cameras(sensors) could be like two different films on 35mm. One concentrating on decent high ISO and overall speed, general use, one for the best resolution. Like having the sharpest ISO25 film and the higher ISO film for occasions that require other kind of performance.

The lower DR might not be in favour, but maybe it could be overcome or maybe I would not care with that 70million pixel camera...

 

[Fred Briggs]

I wouldn't want this ISO 50-200 camera for my birding. But for the
landscapes.. ..YES!

On the other hand just think of the details in the feathers from 70MP, or alternatively the scope for digital zooming by cropping into a small subject and still having reasonable numbers of pixels!

If we decide that full 35mm frame is the size where we want to be
maybe the two cameras(sensors) could be like two different films on
35mm. One concentrating on decent high ISO and overall speed, general
use, one for the best resolution. Like having the sharpest ISO25 film
and the higher ISO film for occasions that require other kind of
performance.

Maybe the day will come when the sensor is a snap in module, and you could buy different ones for different purposes. Oh hang on a minute - we've been there already - it was called film!

Fred

 

[upnorthfar]

Maybe the day will come when the sensor is a snap in module, and you
could buy different ones for different purposes. Oh hang on a minute

• we've been there already - it was called film!

Fred

I don't care if it's an old idea, this is brilliant! =)

 

[John Sheehy]

I agree that there is still opportunity for improvements in category
A below. I have read about new video camera sensors that have one A/D
for each sensor row; this type of architecture has the potential to
reduce A/D noise substantially greatly increasing the D/R of low ISO
images.

Don't get too excited. Lower read noise will never bring shot noise to the levels of low ISO. High ISOs and small sensors will always have significant noise.

--
John

 

[John Sheehy]

You got me thinking this really.
5D sensor area is something like 10 times bigger than G6 digicam with
7,1Mp.
If there was 10 G6 sensors joined side by side to fill the area
instead of the 5D 12Mp sensor...
The resolution advantage would be huge!
And noise.. Wouldn't it be actually better than 12Mp 5D in normal
50-200 sensitivities? Per pixel more yes, but there would be SO many
pixels and the 12Mp is only a tiny crop of the imaginary 71Mp full
size sensor...
And color tones and such.. The huge amount of pixels would perhaps
make a lot smoother tonal changes etc..?

The 5D might have slightly less image noise at ISO 1600, but the 71MP sensor would most likely have less image noise at lower ISOs, and always more resolution (if the optics provide it, of course).

--
John

 

[Don_D]

You got me thinking this really.
5D sensor area is something like 10 times bigger than G6 digicam with
7,1Mp.
If there was 10 G6 sensors joined side by side to fill the area
instead of the 5D 12Mp sensor...
The resolution advantage would be huge!
And noise.. Wouldn't it be actually better than 12Mp 5D in normal
50-200 sensitivities? Per pixel more yes, but there would be SO many
pixels and the 12Mp is only a tiny crop of the imaginary 71Mp full
size sensor...
And color tones and such.. The huge amount of pixels would perhaps
make a lot smoother tonal changes etc..?

Wouldn't you hit the resolution limits of the lens way before you could take advantage of the resolution of such a sensor?

Don
http://www.pbase.com/dond

 

[John Sheehy]

Wouldn't you hit the resolution limits of the lens way before you
could take advantage of the resolution of such a sensor?

What lens are you talking about?

The best telephotos and macros would need hundreds of MPs fullframe to be completely oversampled (especially in the red and blue channels, which have twice the pixel pitch of the overall sensor).

Any lens worth using a 2x TC on is worth having 4x as many pixels; any worth a 2x + 1.4x stacked is worth having 8x as many pixels.
--
John

 

[Don_D]

The best telephotos and macros would need hundreds of MPs fullframe
to be completely oversampled (especially in the red and blue
channels, which have twice the pixel pitch of the overall sensor).

Any lens worth using a 2x TC on is worth having 4x as many pixels;
any worth a 2x + 1.4x stacked is worth having 8x as many pixels.

John,
Thanks for your reply.

After doing a little research (which I should have done before my post) the first reference I found supports your statements. I guess I had a basic misunderstanding in what was limiting the resolution in a DSLR.

However, I am still a bit confused as to why that many lenses are "soft" wide open at the edge of the field using today's DSLRs. Isn't this a lens problem? Wouldn't increasing the pixels on the sensor by 4x be mostly wasted when using such a lens?

Don
http://www.pbase.com/dond

 

[digital_ray_of_light]

I understand what you are trying to compare. I understand that the
experiment you came up with maybe the best available option to
compare pixel areas.

However, what I am trying to say is that the question you are trying
to answer is not answerable by this test, because pixel size is not
independent of sensor size. Pixel size is bound by the sensor size.
Pixels dont exist outside of sensors.

So essentially your experiment is making the assumption that pixels
can be compared independently of the sensor they live on (when
dealing with different size sensors). What I am saying is that pixels
are dependent on the sensor they live on, so they cannot be compared
deterministically in such a fashion.

I hope this is more clear than the previous one

It is clear what you're trying to say, but I don't see any truth to it.

It is not a question of truth to it or not. It is a question of what assumptions were made in the design of the above experiment.

Small pixels from small cameras show us what crops of large sensors
with small pixels could be like. Pixels don't care what role their
playing; they just build up charges.

Certainly the pixels don't care, but they are bound to the sensor. Sensors are designed to deliver the best possible picture at the lowest possible price at their size. Pixels are dependent on that. So this dependence of pixels on the sensor size/design means that you can't compare pixels outside the context of a sensor, thus the experiment does not answer the question asked. Pixels don't exist outside the sensor.

--

Comprehensive 2007 speculation and predictions: http://1001noisycameras.blogspot.com

 

[John Sheehy]

Certainly the pixels don't care, but they are bound to the sensor.
Sensors are designed to deliver the best possible picture at the
lowest possible price at their size. Pixels are dependent on that. So
this dependence of pixels on the sensor size/design means that you
can't compare pixels outside the context of a sensor, thus the
experiment does not answer the question asked. Pixels don't exist
outside the sensor.

I can't find any sense in what you say, or see the point of your perspective. You sound like an apologist for industrial lethargy. Big pixels are a compromise, not an ideal. We can see that by comparing crops of the same physical size on the sensor. What don't you understand about this?

--
John

 

[John Sheehy]

After doing a little research (which I should have done before my
post) the first reference I found supports your statements. I guess
I had a basic misunderstanding in what was limiting the resolution in
a DSLR.

Well, a lot of the time it's the photographer. You need to achieve focus and stability, and freeze the subject if necessary, and stay away from diffractive f-stops to make maximum use of the lens' sharpess.

However, I am still a bit confused as to why that many lenses are
"soft" wide open at the edge of the field using today's DSLRs. Isn't
this a lens problem? Wouldn't increasing the pixels on the sensor by
4x be mostly wasted when using such a lens?

Do we want to limit all cameras by the fact that some lenses don't deliver? Most of the comments about how systems are limited by the lenses come from people who shoot larger sensors with wide-angle lenses, and medium-to-wide zooms. They are not the sharpest lenses. Even L lenses like the 16-35 are not really very sharp.

Even if a lens doesn't deliver at the edges, it may still deliver in the center. Would you wish that all your lenses became as soft in the center as they are in the corners? With more resolution, you have better options for correcting aberrations in software, as well. You also lose less detial when doing perspective corrections and rotations when you are already over-sampling the lens. Over-sampling isn't a total waste.
--
John