dpreview pixel density metric

In article , Kennedy McEwen
wrote:

Specifying
something as generic as pixel density, which has to cope with all the
possible sensor configurations isn't as straight forward as it initially
appears. Even area assessment has its limitations - for example, have
they really got the Foveon / Sigma numbers right?

it looks like the pixel density is correct, however, that doesn't
convey the additional noise inherent to the foveon chip.

Either way they
represent it (ie. taking account of the 3 stacked pixels in each
position or not) someone will use it the wrong way and reach false
conclusions.

something which the sigma fans love to do.

Alan Browne wrote:

I just noticed that dpreview have finally added a pixel density metric
which could save some time when comparing cameras.

However they express it as pixels / cm^2.

I just wrote them a spot of feedback recommending that they do it as
pixel pitch in lp/mm which is more consistent with lens performance
measurements.

(Even if they did it as pixel pitch in pixels/mm, it would be easier to
imagine the meaning than pixels/cm^2)

It doesn't matter how they express it - it _will_ lead to the erroneous
conclusion that lower pixel density alone offers some advantage - other
than smaller files size. (It only offers an advantage in noise if sensor
efficiency falls as pixel density increases - and that hasn't generally
been the case with DSLR sensor development)

For anyone bothering to work out diffraction effects or lens MTF based
on pixel density, then calculating "pixel pitch" is a triviality -
DPReview's figure can't help much.

For P&S cameras it's a waste of time, and lens performance data isn't
known. If DPReview wanted to throw in a measure, then perhaps if they'd
used f-stop at which airy disk diameter exceeds 2x pixel pitch, it might
have been of some use as a time saver.

In article , Alan Browne
writes
Kennedy McEwen wrote:

I guess they are using an area measurement to cope with anisotropic
oddities like the Nikon D1x rectangular pixels (and hence two
different linear measures) or the Fuji octagonal and mixed size
pixels. They could just use the square root as pixels/mm but then
someone would point out that they weren't using the right numbers for
such cases. Specifying something as generic as pixel density, which
has to cope with all the possible sensor configurations isn't as
straight forward as it initially appears. Even area assessment has
its limitations - for example, have they really got the Foveon /
Sigma numbers right? Either way they represent it (ie. taking
account of the 3 stacked pixels in each position or not) someone will
use it the wrong way and reach false conclusions.

I beg to differ. First of all they're showing the metric to 3
significant digits, so that would filter out the fine resolution
resulting from what you suggest.

Why do you think the number of significant figures makes any difference
with anisotropic detectors? The Nikon D1x, for example, only needs ONE
significant digit to differentiate between the vertical and horizontal
pixel densities. Similarly the Sigma cameras need no more than one
significant digit to differentiate between their actual resolution and
their pixel density.

What matters is really how well a sensor matches up to a lens, IMO, as
that determined a large part of the capture quality.

No, that's not what matters at all. That's just what you are trying to
misuse the parameter for. Their reason for specifying the parameter in
this way was given in the Dpreview explanation of it, and it is a not
related to lens performance at all. If you want to relate sensor
resolution to lens performance then you need another metric altogether,
and not geometric pixel density.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

In article , nospam
writes
In article , Kennedy McEwen
wrote:

Specifying
something as generic as pixel density, which has to cope with all the
possible sensor configurations isn't as straight forward as it initially
appears. Even area assessment has its limitations - for example, have
they really got the Foveon / Sigma numbers right?

it looks like the pixel density is correct, however, that doesn't
convey the additional noise inherent to the foveon chip.

Nor does it take account of the extra two photodiodes in each pixel
area.

Either way they
represent it (ie. taking account of the 3 stacked pixels in each
position or not) someone will use it the wrong way and reach false
conclusions.

something which the sigma fans love to do.

as do the bayer fans...

Only one way to satisfy all of the people all of the time...
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

Alan Browne wrote:
[]
Not really, but it could be simply pixels per mm if lp/mm drives you
crazy. (Lens MTF remains as lp/mm however so you'd have to divede
px/mm by two to get line pairs...

We used to use cycles per mm, rather than line pairs, mainly in reference
to the sinusoidal waveforms used throughout the rest of the system.....

Cheers,
David

In article , Kennedy McEwen
wrote:

Specifying
something as generic as pixel density, which has to cope with all the
possible sensor configurations isn't as straight forward as it initially
appears. Even area assessment has its limitations - for example, have
they really got the Foveon / Sigma numbers right?

it looks like the pixel density is correct, however, that doesn't
convey the additional noise inherent to the foveon chip.

Nor does it take account of the extra two photodiodes in each pixel
area.

which is a major factor that causes the additional noise that i
mentioned. since they are noisier pixels, the number is a little
misleading.

Either way they
represent it (ie. taking account of the 3 stacked pixels in each
position or not) someone will use it the wrong way and reach false
conclusions.

something which the sigma fans love to do.

as do the bayer fans...

bayer fans don't artificially inflate the pixel count to make the
camera appear to be better than it really is.

In article , David J
Taylor
writes
Alan Browne wrote:
[]
Not really, but it could be simply pixels per mm if lp/mm drives you
crazy. (Lens MTF remains as lp/mm however so you'd have to divede
px/mm by two to get line pairs...

We used to use cycles per mm, rather than line pairs, mainly in reference
to the sinusoidal waveforms used throughout the rest of the system.....

and we still should, for accurate work. The line pair is only
equivalent to a cycle close to the resolution limit of a system, where
the higher harmonics are not resolved. Unfortunately, with typical
digital sensors, that is not usually the case.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

In article , nospam
writes
In article , Kennedy McEwen
wrote:

Specifying
something as generic as pixel density, which has to cope with all the
possible sensor configurations isn't as straight forward as it initially
appears. Even area assessment has its limitations - for example, have
they really got the Foveon / Sigma numbers right?

it looks like the pixel density is correct, however, that doesn't
convey the additional noise inherent to the foveon chip.

Nor does it take account of the extra two photodiodes in each pixel
area.

which is a major factor that causes the additional noise that i
mentioned.

No it isn't, in fact, as you seem to wish to delve to the in squalor of
pedantry, the additional photodiodes are not any noisier at all!

It is the relatively poor colour filtration of each diode which requires
much heavier matrix manipulation of the diode output to create RGB data
that results in more noise. It also means that the colour purity of the
data is more prone to error.

since they are noisier pixels, the number is a little
misleading.

Either way they
represent it (ie. taking account of the 3 stacked pixels in each
position or not) someone will use it the wrong way and reach false
conclusions.

something which the sigma fans love to do.

as do the bayer fans...

bayer fans don't artificially inflate the pixel count to make the
camera appear to be better than it really is.

The question isn't one of exaggerating pixel count per se, neither group
actually does that and both claim the true pixel count of their own
favoured technology.

The question is about what pixel count is relevant to image resolution.
The Bayer camp are certainly as guilty of exaggerating that as much as
the Foveon camp.

What do you think the discussions about hard and soft AA filters are all
about? There is no value whatsoever to a high Bayer pixel count if it
is all filtered away at the AA stage, or in the optic itself for that
matter. That is what ultimately what makes the pixel density metric
quite meaningless in terms of image quality. It is only meaningful as a
measure of how tightly the pixels are packed, not how useful they are.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

In article , Kennedy McEwen
wrote:

In article , nospam
writes
In article , Kennedy McEwen
wrote:

Specifying
something as generic as pixel density, which has to cope with all the
possible sensor configurations isn't as straight forward as it initially
appears. Even area assessment has its limitations - for example, have
they really got the Foveon / Sigma numbers right?

it looks like the pixel density is correct, however, that doesn't
convey the additional noise inherent to the foveon chip.

Nor does it take account of the extra two photodiodes in each pixel
area.

which is a major factor that causes the additional noise that i
mentioned.

No it isn't, in fact, as you seem to wish to delve to the in squalor of
pedantry, the additional photodiodes are not any noisier at all!

by slicing the pixel into three layers, each layer will have a lower
well capacity (particularly the top layer since they're not equivalent
thicknesses).

so instead of (say) 60k photons for the entire pixel, it would be 20k
each (for equivalent size slices) or for foveon, ~8k for the top layer
(it is about 1/8th the thickness of the total) which is quite low.
looking at foveon images, the blue channel (mostly from the top layer)
is quite noisy.

It is the relatively poor colour filtration of each diode which requires
much heavier matrix manipulation of the diode output to create RGB data
that results in more noise. It also means that the colour purity of the
data is more prone to error.

that's the other major factor that makes the output noisier and subject
to weird colour casts, metamerism, etc.

Either way they
represent it (ie. taking account of the 3 stacked pixels in each
position or not) someone will use it the wrong way and reach false
conclusions.

something which the sigma fans love to do.

as do the bayer fans...

bayer fans don't artificially inflate the pixel count to make the
camera appear to be better than it really is.

The question isn't one of exaggerating pixel count per se, neither group
actually does that and both claim the true pixel count of their own
favoured technology.

only sigma/foveon attempt to redefine the term (and sigma uses it
inconsistently, further proof that they're trying to deceive).

The question is about what pixel count is relevant to image resolution.
The Bayer camp are certainly as guilty of exaggerating that as much as
the Foveon camp.

they're not guilty at all. bayer uses the term 'pixel' correctly, as
it has been used long before there *was* a bayer or foveon.

sigma even used 'pixel' correctly for the sd9, and then decided to
multiply the pixel count by 3 for the sd10, despite the fact the sensor
was identical (other than microlenses).

What do you think the discussions about hard and soft AA filters are all
about? There is no value whatsoever to a high Bayer pixel count if it
is all filtered away at the AA stage, or in the optic itself for that
matter. That is what ultimately what makes the pixel density metric
quite meaningless in terms of image quality. It is only meaningful as a
measure of how tightly the pixels are packed, not how useful they are.

the aa filter limits detail that can't be resolved. remove it (or use
a weak one) and you get alias artifacts. apparently some people
(namely sigma fans) like the look and think it's additional detail, but
the chip isn't resolving real detail at all.

On Sun, 06 Jul 2008 05:39:38 GMT, David J Taylor wrote:

I agree that per mm^2 might be a better unit, but please /NOT/ the
"pairs" again! Totally ambiguous!
. . .
On reflection, although I don't like the "pairs" ambiguity, I do think
that a linear measure rather than an area measure would be a better
representation of performance. Pixel spacing (in micrometers) would
appear to achieve this.

No matter which definition is used, some will cheer, others will
jeer. "To double a 6mp camera's resolution, get one with a 12mp
sensor". "Nonsense, to double the resolution you'd need 24mp".

In the end it doesn't matter much. Common usage/conventional
wisdom will rule the day, whether right, wrong or silly. At least,
it will provide fodder for discussion here at least a step or two
above the level of a typical RichA/Rita/Arthurian troll. If you
were twice as hungry as usual, would you buy a pizza with twice the
area or twice the diameter?

My answer - I'd buy the usual size pizza, since there'd still be
slices left over.