Nikon D800/E $3000, cheaper than I thought

On Thu, 1 Mar 2012 23:12:40 +0100, Wolfgang Weisselberg
wrote:
: John A wrote:
: On Tue, 07 Feb 2012 08:23:00 +0000, Bruce
:
: It will be interesting to see if Nikon can offer an AA filter removal
: service for D800 customers who subsequently see the light. ;-)
:
: You would think they would have charged more for the version with
: the filter, if so many more ignorantly believe it's better with it
: than without.
:
: Some think they get added value from no AA filter. Let them.
: Let them be gouged for getting an essentially worse/niche product.
:
: Sweet deal for Nikon, too: bigger market, extra money from lunatics
: and practically no extra work to produce the special version.

If I understand correctly (and I admit that it's likely I don't), the function
of the AA filter is to statistically degrade the RAW image, to prevent it from
fooling the eye into seeing patterns that are not really there. So how is that
different from applying an algorithm with a similar effect to the RAW image in
post-processing (or in the camera)? IOW, why can't there be a software or
firmware AA filter? I.e., why do we need the D800E at all? (Actually, I don't,
of course; I'm a Canonian. We have our own problems just now, but whether to
opt for the "E" version of the D800 isn't one of them.)

Bob

On 3/03/2012 11:41 a.m., Robert Coe wrote:
On Thu, 1 Mar 2012 23:12:40 +0100, Wolfgang Weisselberg
wrote:
: John wrote:
: On Tue, 07 Feb 2012 08:23:00 +0000,
:
:It will be interesting to see if Nikon can offer an AA filter removal
:service for D800 customers who subsequently see the light. ;-)
:
: You would think they would have charged more for the version with
: the filter, if so many more ignorantly believe it's better with it
: than without.
:
: Some think they get added value from no AA filter. Let them.
: Let them be gouged for getting an essentially worse/niche product.
:
: Sweet deal for Nikon, too: bigger market, extra money from lunatics
: and practically no extra work to produce the special version.

If I understand correctly (and I admit that it's likely I don't), the function
of the AA filter is to statistically degrade the RAW image, to prevent it from
fooling the eye into seeing patterns that are not really there.
Not quite. The eyes aren't fooled at all - the sensor is - and the
result which is seen by the eyes is 100% real.
So how is that
different from applying an algorithm with a similar effect to the RAW image in
post-processing (or in the camera)? IOW, why can't there be a software or
firmware AA filter? I.e., why do we need the D800E at all? (Actually, I don't,
of course; I'm a Canonian. We have our own problems just now, but whether to
opt for the "E" version of the D800 isn't one of them.)

To put it simply, software can deal with one aspect of having no AA
filter quite well. Colour artifacts colour moire can be removed, at the
expense of some colour resolution - which isn't usually very important
visually anyway as our eyes have far greater resolution of luminance
than colour.
But software can't "decide", after the photo has been taken, whether
remaining aliasing and "luminance" banding in an image was present in
the scene, or if it wasn't. There's no practical way to remove it after
the fact.

Robert Coe writes:

If I understand correctly (and I admit that it's likely I don't), the function
of the AA filter is to statistically degrade the RAW image, to prevent it from
fooling the eye into seeing patterns that are not really there.

No.

The AA filter prevents interactions between regular subject patterns in
certain frequency ranges (up near the limit of the system) and the
regular grid of the sensor. (That's why it wasn't a problem with film;
the film grain wasn't arranged in a regular grid.)

Once the alias artifacts are recorded by the sensor, it's extremely
difficult to remove, because it's up right near the limiting resolution
of the system (meaning very small bits to edit one at a time).
--
David Dyer-Bennet, ; http://dd-b.net/
Snapshots: http://dd-b.net/dd-b/SnapshotAlbum/data/
Photos: http://dd-b.net/photography/gallery/
Dragaera: http://dragaera.info

David Dyer-Bennet wrote:
Robert Coe writes:

If I understand correctly (and I admit that it's likely I don't), the function
of the AA filter is to statistically degrade the RAW image, to prevent it from
fooling the eye into seeing patterns that are not really there.

No.

The AA filter prevents interactions between regular subject patterns in
certain frequency ranges (up near the limit of the system) and the
regular grid of the sensor. (That's why it wasn't a problem with film;
the film grain wasn't arranged in a regular grid.)

Once the alias artifacts are recorded by the sensor, it's extremely
difficult to remove, because it's up right near the limiting resolution
of the system (meaning very small bits to edit one at a time).

That's close, but a bit misleading too.

The D800E, with no filter, will have improved high
frequency response and a better signal to noise ratio
"up right near the limiting resolution of the system"
(just below the Nyquist Limit). But it will have
aliasing distortion artifacts spread through out the
spectrum below the Nyquist Limit, including in some
images moiré.

The D800, with an AA filter, will not quite match the
response just below the Nyquist Limit. That can
probably be corrected, as far as signal level goes, with
High Pass Sharpening... but that will also increase the
noise at those frequencies, so the SNR will remain lower
than the D800E in the bandwidth just below the Nyquist
Limit. But the D800 will not have anything like the
amount of aliasing distortion at frequencies lower than
that.

From that we can consider where the two will differ in
actual use. For prints made at the original pixel
dimension, 7360 pixels across horizontally, which for
printing on an Epson printer at 360 PPI amounts to
7360/360 = 20.4 inches, or larger (with resampling to a
greater pixel dimension), the D800E will provide a
slight improvement in high frequency detail just below
the Nyquist Limit, while the D800 will be cleaner at
lower frequencies. Which one is best is a judgment call
that depends on the specific image and the specific use.

But...

For any image printed at less than 20.4 inches, which is
to say that it must be resampled to a smaller pixel
dimensions, the low pass filter effect of resampling
will *absolutely* remove the extra resolution near the
Nyquist Limit for the sensor from a D800E image, but
that will not change the aliasing distortion at lower
frequencies. Hence it is not a judgment call and does
not depend on the image or the use, the D800 will
produce an equal or better print at any size less than
20 inches!

Other parameters might also affect the benefits
available from the D800E. For example, most of the
prints I make are indeed at 16x20 and larger and hence
it would at first blush appear that my work might
possibly benefit from a D800E over the D800. Except
that I print about 90% of what I do on canvas! My
widest printer is a 24 inch Epson 7890, and with canvas
I doubt it can benefit from the extra detail a D800E
produces unless I start doing six foot long panoramas!

--
Floyd L. Davidson http://www.apaflo.com/
Ukpeagvik (Barrow, Alaska)

Robert Coe wrote:
On Thu, 1 Mar 2012 23:12:40 +0100, Wolfgang Weisselberg
wrote:
: John A wrote:
: On Tue, 07 Feb 2012 08:23:00 +0000, Bruce
:
: It will be interesting to see if Nikon can offer an AA filter removal
: service for D800 customers who subsequently see the light. ;-)
:
: You would think they would have charged more for the version with
: the filter, if so many more ignorantly believe it's better with it
: than without.
:
: Some think they get added value from no AA filter. Let them.
: Let them be gouged for getting an essentially worse/niche product.
:
: Sweet deal for Nikon, too: bigger market, extra money from lunatics
: and practically no extra work to produce the special version.

If I understand correctly (and I admit that it's likely I don't), the function
of the AA filter is to statistically degrade the RAW image, to prevent it from
fooling the eye into seeing patterns that are not really there.

Actually, it's not "fooling" the eye. It's the sensor
itself. The sensor itself can 'imagine' things (like moire) that are
not really there. That false information ends up in the RAW.
Once there (and in the JPEG) it's real, not a trick of the
eye.

So how is that
different from applying an algorithm with a similar effect to the RAW image in
post-processing (or in the camera)? IOW, why can't there be a software or
firmware AA filter?

If the sensor has to digitize a pattern that has a higher
frequency than the sensor can resolve (e.g. more than 1 line
pair per pixel pair) that causes a pattern that's frequency wise
mirrored, i.e. a lower frequency.

The same happens when you overlay 2 pairs of lines or grids
at an angle, see here
http://upload.wikimedia.org/wikipedi...bb/Moire02.gif
how there are created horizontal dark and bright stripes,
and how you get completely wrong patterns here on these
parrot feathers (caused by the sensor)
http://upload.wikimedia.org/wikipedi...t_feathers.jpg
or here, 2 fences (this moire is not caused by the sensor)
http://www.ics.uci.edu/~eppstein/pix...nceMoire-m.jpg
And this wall isn't build wavy:
http://upload.wikimedia.org/wikipedi...icks_small.jpg

It's not a trick of the eyes, you can zoom in or measure the
spots with the pipette tool ...

.... it's in the data.
And because it's in the data, there's no chance in hell for
an algorithm to detect if that low or mid frequency pattern is
real (maybe you copied a photo, maybe it's 2 fences and real) or
if it *only* happened due to the restricted resolution of the
sensor.

(The same problem applies to downsizing the image; if not
done well you can get moire ...
http://bvdwolf.home.xs4all.nl/main/f...own_sample.htm
)

Actually, you have the exact same problem in audio, so you
use a low pass filter cutting away everything too close to the
sample rate (say, everything above 20kHz in an audio CD
(sample rate 44.1kHz, so it could at best resolve a 22.05kHz
frequency). The 2 fences in audio is 2 audio frequencies
close to each other ... you get the difference as beat
frequency, even though that beat is never played. (In fact,
a trick of the ear is binaural beats, where the ears each get
a slightly different frequency and the difference beat is
actually 'heard':
http://en.wikipedia.org/wiki/Binaural_beats

Try it yourself: draw a long sine wave and mark out points at
a distance of a 3/4th wave. Plot these points and draw the
lowest frequency sine wave that fits through them ... it's
not your original wave.

I.e., why do we need the D800E at all?

Some people (who already "hot rod" 5Ds and 5DIIs claim
there's some resolution increase ... I guess on the order of
30%.
http://maxmax.com/hot_rod_visible.htm

Of course, to be fair you'd need to apply more sharpening to the
AA filter variant. I have adjusted the exposure and upped
the sharpness of the 'stock' variant somewhat (and gave both
a touch of lens correction), getting this (upsampled to 200%
in each dimension):
http://www.smugmug.com/gallery/21744124_pt7FN6

Result: The 5D's AA filter is too weak already, and you only get
a tiny bit of extra resolution for quite a bit of problems.
I *don't* like my thin black details in colour.

-Wolfgang

In article , David Dyer-Bennet
wrote:

Once the alias artifacts are recorded by the sensor, it's extremely
difficult to remove, because it's up right near the limiting resolution
of the system (meaning very small bits to edit one at a time).

so extremely difficult, that it's actually impossible to remove after
it's sampled. there's no way to know whether something is an alias
artifact or actual detail.

In article , Robert Coe
wrote:

If I understand correctly (and I admit that it's likely I don't), the function
of the AA filter is to statistically degrade the RAW image, to prevent it from
fooling the eye into seeing patterns that are not really there.

definitely not. basically, every sensor has an upper limit of the
amount of detail it can resolve and anything finer than that will
alias, or create false details. an anti-alias filter blocks detail
higher than the capabilities of the sensor so that there is no false
detail. nothing is perfect so there's a tradeoff as to how strong an
anti-alias filter should be.

So how is that
different from applying an algorithm with a similar effect to the RAW image in
post-processing (or in the camera)? IOW, why can't there be a software or
firmware AA filter?

it's different because once you capture the image, you have *no* way to
tell whether something is real detail or an alias artifact, so you
can't do it afterwards in post. you *must* filter it before capture.

I.e., why do we need the D800E at all? (Actually, I don't,
of course; I'm a Canonian. We have our own problems just now, but whether to
opt for the "E" version of the D800 isn't one of them.)

the d800e is mainly for those who don't understand sampling theory and
think no aa filter is a good idea.

nospam writes:

In article , David Dyer-Bennet
wrote:

Once the alias artifacts are recorded by the sensor, it's extremely
difficult to remove, because it's up right near the limiting resolution
of the system (meaning very small bits to edit one at a time).

so extremely difficult, that it's actually impossible to remove after
it's sampled. there's no way to know whether something is an alias
artifact or actual detail.

That's relevant for certain kinds of scientific photography, but not for
artistic photography. Different kinds of documentary photography fall
somewhere in between. For nearly all portraits, what matters is that it
"looks right", not that it *is* right. And one can often (always, with
enough work) repair the moire so that it looks right. (Proof: people
can paint photo-realistic paintings from scratch; therefore they can, if
they want to badly enough, replace the areas damaged by moire or
whatever from scratch, if no lesser fix is satisfactory.)
--
David Dyer-Bennet, ; http://dd-b.net/
Snapshots: http://dd-b.net/dd-b/SnapshotAlbum/data/
Photos: http://dd-b.net/photography/gallery/
Dragaera: http://dragaera.info

In article , David Dyer-Bennet
wrote:

Once the alias artifacts are recorded by the sensor, it's extremely
difficult to remove, because it's up right near the limiting resolution
of the system (meaning very small bits to edit one at a time).

so extremely difficult, that it's actually impossible to remove after
it's sampled. there's no way to know whether something is an alias
artifact or actual detail.

That's relevant for certain kinds of scientific photography, but not for
artistic photography. Different kinds of documentary photography fall
somewhere in between. For nearly all portraits, what matters is that it
"looks right", not that it *is* right.

it's relevant for all photography. aliasing means it won't look right
because there are details in the photo that *weren't* in the subject.

some people are fooled by that (sigma/foveon users in particular), so
they want the artifacts to stay.

And one can often (always, with
enough work) repair the moire so that it looks right. (Proof: people
can paint photo-realistic paintings from scratch; therefore they can, if
they want to badly enough, replace the areas damaged by moire or
whatever from scratch, if no lesser fix is satisfactory.)

that's one of the most ridiculous things i've read in ages. why even
use a camera at all you are going to repaint the whole image from
scratch because it's full of artifacts?

nospam writes:

In article , David Dyer-Bennet
wrote:

Once the alias artifacts are recorded by the sensor, it's extremely
difficult to remove, because it's up right near the limiting resolution
of the system (meaning very small bits to edit one at a time).

so extremely difficult, that it's actually impossible to remove after
it's sampled. there's no way to know whether something is an alias
artifact or actual detail.

That's relevant for certain kinds of scientific photography, but not for
artistic photography. Different kinds of documentary photography fall
somewhere in between. For nearly all portraits, what matters is that it
"looks right", not that it *is* right.

it's relevant for all photography. aliasing means it won't look right
because there are details in the photo that *weren't* in the subject.

That's incorrect. Some blatant situations like moire will certainly
look wrong, but aliasing that DOESN'T build up global patterns like that
will largely go unnoticed.

some people are fooled by that (sigma/foveon users in particular), so
they want the artifacts to stay.

And users of most medium-format digital bakcs, apparently.

And one can often (always, with
enough work) repair the moire so that it looks right. (Proof: people
can paint photo-realistic paintings from scratch; therefore they can, if
they want to badly enough, replace the areas damaged by moire or
whatever from scratch, if no lesser fix is satisfactory.)

that's one of the most ridiculous things i've read in ages. why even
use a camera at all you are going to repaint the whole image from
scratch because it's full of artifacts?

Because actual visible artifacts are very very rare; so most of the time
you win.
--
David Dyer-Bennet, ; http://dd-b.net/
Snapshots: http://dd-b.net/dd-b/SnapshotAlbum/data/
Photos: http://dd-b.net/photography/gallery/
Dragaera: http://dragaera.info