Pentax K10D beats (sharpness, detail) Canon 40D

David J Taylor wrote:
Wolfgang Weisselberg wrote:
David J Taylor
wrote:
Floyd L. Davidson wrote:

But a 43.1 KHz tone is extremely easy to filter out.

Of course, I was trying to make the numbers easier, and to
demonstrate how something could alias right down into the wanted
band.

But a similar image frequency is far harder to filter out.

As I already mentioned, in audio oversampling is typically used to
ameliorate these issues, but as far as I know, oversampling has not
yet been applied to DSLRs.

Oversampling in audio, unless I misunderstand, would be sampling
with a much higher frequency than the target 44.1kHz frequency.

In DSLRs that would mean using 4 (2x oversampling) or 16 (4x
oversampling) pixels where there is now one. For a 6MP resolution
you'd need 24 million pixels at just 2x oversampling in width
and height. For physical reasons so many pixels need large (and
expensive) sensors or must deal with small full well sizes and
photon noise.

On the other hand, the new sRAWs may be downsampled from ordinary
RAWs, and thus count as oversampling + downsampling. Of course
you can get the same effect by resizing your photo intelligently
from 8MPix to, say 2MPix (2x) or 0.9MPix (3x)

-Wolfgang

Wolfgang,

Yes, the oversampling allows the first (analog) filter to be simpler, and
the subsequent filtering to be done digitally. Final samples are
delivered at 44.1KHz (or higher in studio work).

If oversampling were used for digital cameras, I don't think that the
photon noise and dynamic range would be significantly worse, as the same
area of silicon is used per output pixel. In your 2Mpix or 0.9Mpix
analogy, it would mean that you could use an anti-aliasing filter with an
equivalently strong cut-off (i.e. less sharp 8Mpix images).

There remains one problem with digital photography though, as someone
mentioned recently, that the very simple anti-aliasing filter used does
not have a sharp cut-off just before the Nyquist frequency, it has a more
gentle slope, meaning that there will be some compromise between what some
call "resolution" and the damaging effects of aliasing. Different cameras
will behave differently, and different users may prefer different results.

That was the point I was trying to get at in my original
question. Many DSLR users already accept that different images will
respond differently to different levels of sharpening, and that for
best effect sharpening should be applied at whatever level of
resolution the final print or screen image is to be viewed at, which
may not be the same as the original camera image resolution.

Hence most cameras now come with menu-settable changes in the degree
of sharpening applied to their jpgs, and to what level of image
resolution. Since the extra power and sophistication of large computer
processors compared to those in the camera mean that image editors can
employ more sophisticated sharpening than the cameras, many DSLR users
now prefer to do the minimum in camera and most of it later in their
own preferred image editor.

What I don't understand is why the same kind of flexibility is not
applied to AA filtering. I can't see any technological barriers to
it. Given the current state of camera technology it would only be of
benefit to those using expensive high quality kit, but the fact that
some camera makers (such as Leica, Sony, & Fuji, IIRC) are already in
their top models edging in the direction of less optical AA filtering
in order to exploit more of the native resolution in their technology
it seems to me that camera technology has now got to the point where
this has become relevant.

--
Chris Malcolm DoD #205
IPAB, Informatics, JCMB, King's Buildings, Edinburgh, EH9 3JZ, UK
[http://www.dai.ed.ac.uk/homes/cam/]

Chris Malcolm wrote:
David J Taylor
wrote:
Floyd L. Davidson wrote:

But a 43.1 KHz tone is extremely easy to filter out.

Of course, I was trying to make the numbers easier, and to
demonstrate how something could alias right down into the wanted
band.

Filters are complex only when the cutoff has a steep
slope.

.. which is required for audio as an essentially flat response is
required, with a sharp cut-off before the Nyquist frequency. Making
the pass-band acceptably flat also increases the complexity of the
filter, as does control of group delay. As I already mentioned, in
audio oversampling is typically used to ameliorate these issues, but
as far as I know, oversampling has not yet been applied to DSLRs.

Because in the case of digital cameras you'd need a lens capable of
resolution at the oversampling frequencies, and with today's
technology that would cost far too much to be worth it as a component
of an anti-aliasing system. That's why in the realm of digital cameras
we're still in the engineering morass of systems which are struggling
to reach the levels of quality at which the practical problems will be
as well behaved as the theoretical mathematical analyses.

The resolution of the lens would remain unchanged. You are simply trying
to capture a more accurate "8Mpix" (or whatever) representation of the
image in the focal plane of the lens. I agree that the cost would be too
high at the moment, and with your comments on the subsequent results!

Cheers,
David

John Bean wrote:
[]
Genuine question (I really don't know): does the microphone
need to have a response in excess of the oversampling
frequency when making a digital audio recording?

It seems to me that the lens and microphone have an
analogous similarity in this discussion.

No, it does not.

David

Chris Malcolm wrote:
[]
I know. I used to be one of them. More than twenty five years ago I
was writing code to get rid of aliasing artefacts in monochrome
256x256 digital camera images :-) That was back in the old days when
the only way to acquire a digital camera was to make it yourself by
unsoldering the metal top of a military spec dynamic RAM chip and then
focusing an image on it with a lens.

Fascinating - did you have a lot of success with your code? What nature
of test images did you use (edges or bar wedges), and did you end up
designing in the frequency or time/space domain?

Cheers,
David

On Oct 28, 1:14 pm, Chris Malcolm wrote:
John Bean wrote:
On 27 Oct 2007 09:27:51 GMT, Chris Malcolm
wrote:
I can't understand why you say that no amount of post-processing will
remove it without degrading the image. I don't understand why you
can't do the same post-processing computationally as you can
optically. I see no mathematical or computational barrier to doing
exactly the same thing computationally as optically. I thought it was
just a question of convenience and marketing.
Because there's no way to separate what is real from what is
fake, that's the whole problem.

That's not what I was talking about. I was simply talking about the
theoretical equivalence of an optical AA filter, and a computational
one designed to do exactly the same job.

How can it do the same job? Once you have sampled the signal, it has
aliasing artifacts; it's not really possible to tell which are
aliasing artifacts and which detail (but I imagine there are heuristic
algorithms; even the best I've seen seem to just remove the colour
artifacts only).



The AA filter isn't there to
"mask out" false detail, it's there to prevent it from
happening by eliminating the out of band spacial frequencies
that would interfere with the pitch of the sensor grid to
produce the aliasing in the first place.

That *is* what I was talking about, and my point was that that problem
only occurs in certain images. So if you apply the necessary filtering
to the images that need it, you can get higher detail resolution in
those that don't. In fact few camera makers apply enough AA filtering
to get rid of *all* the problems *all* the time. They apply enough to
get rid of most of it. "Most" is a question of taste. As someone else
pointed out, the Leica M8 designers preferred a weaker AA filter and
more detail. Which of course led certain ignorant reviewers to
pontificate about there being a fault in the camera.

On Oct 28, 1:26 pm, "David J Taylor" -this-
bit.nor-this-bit.co.uk wrote:
Chris Malcolm wrote:

[]

That's not what I was talking about. I was simply talking about the
theoretical equivalence of an optical AA filter, and a computational
one designed to do exactly the same job.

Chris,

One problem with filtering in the optical domain may be that in the
computational domain, negative values of light intensity are allowed,
whereas they don't exist for broadband incoherent light detected by a
silicon sensor.

Are they possible for monochromatic and/or coherent light?

So in the optical domain you are constrained to a
positive impulse response for the filter (I think, corrections welcome!).


One thing I never understood is that people keep talking about ideal
filters as step functions. But such a filter would cause ringing at
the cutoff frequency); in fact, there is always a tradeoff between how
much power you allow to pass in frequencies higher than your "cutoff"
and how much signal distortion you get back in real space. I've seen
several papers discussing this (mainly by introducing different
metrics and quantifying several different filter shapes).

So here's my question: you seem to know about such filters in audio.
What is done there? I know that oversampling is common, but do people
normally strive for sharp cutoffs? Is this tradeoff important there?

Of course, this doesn't alter the fact you are trying to prevent the
higher spatial frequencies from reaching the sensor before the spatial
sampling.

Whether these frequencies are present does indeed depend on the image and
other factors, and how much degradation of the image by artefacts or by
lack of the higher spatial frequencies an individual will accept is a
subjective measure.

On Sun, 28 Oct 2007 11:33:14 GMT, "David J Taylor"

wrote:

John Bean wrote:
[]
Genuine question (I really don't know): does the microphone
need to have a response in excess of the oversampling
frequency when making a digital audio recording?

It seems to me that the lens and microphone have an
analogous similarity in this discussion.

No, it does not.


I'd have been surprised if it did, but since practical
digital audio (rather than theory) is not within my area of
expertise I thought I'd ask.

Far too many people around here like to sound like experts
whether or not they actually are.

--
John Bean

On Oct 28, 1:46 pm, Chris Malcolm wrote:
David J Taylor wrote:



Chris Malcolm wrote:
David J Taylor
No, it's false sharpness. Once the higher spatial frequencies have
been aliased down to lower, more visible frequencies, no amount of
post-processing will remove the damage done without causing serious
degradation of the image.

This is what I can't understand. Aliasing is a particular problem
which occurs with a particular kind of image detail. You can't get rid
of it, oprtically or by image processing, without losing some of the
image detail and resolution that would be present in the image without
the filtering. It's a trade off. To get rid of all possible aliasing
you have to lose quite a bit of detail. That's why most camera
manufacturers don't get rid of all of it, they just get rid of most
of it, and where "most" lies in the trade off between aliasing
artefacts and detail loss is a question of taste.
Yes, in optics it's a compromise, as optical anti-aliasing filters are
much simpler than those used in audio. How much of a problem it is also
depends on the subject (how much high-frequency detail) and the lens (what
is the lens MTF around and above the Nyquist frequency). A long exposure
through turbulent atmosphere can also affect the MTF between subject and
lens.
I can't understand why you say that no amount of post-processing will
remove it without degrading the image. I don't understand why you
can't do the same post-processing computationally as you can
optically. I see no mathematical or computational barrier to doing
exactly the same thing computationally as optically. I thought it was
just a question of convenience and marketing.
Perhaps it's easier to think in audio terms. With inadequate filtering
before digitisation, a CD recording system sampling at 44.1KHz would
render a signal which had an actual frequency of 43.1KHz as a tone of
1KHz. Of course, an audio system will have a very good low-pass filter
before the digitisation, ensure that no 43.1KHz information reaches the
ADC. But once you have the 1KHz present, the only way to remove it is to
remove /any/ 1KHz information in the signal. Is that any clearer?

I understand anti-aliasing perfectly in audio. What makes the audio
problem and its typical solutions different is that it's easy with
quite cheap technology to make systems with accurate responses well
above the frequency range of human hearing. So the aliasing problems
and the technology to remove them behave very much more like the
simplified theoretical mathematical models. That makes the practical
engineering and design problems much simpler.


Well, is there some aspect of the discussion here that you think
doesn't really correspond to reality? (since you mention simplified
theoretical models, which is of course true). I am curious, rather
than confrontational.



BTW: the design of the low-pass filter is complex and can cause overshoots
and ringing as seen in some photos. Multi-level sampling at different
rates can be used to simplify the filter design. It's quite an
interesting area - there's probably quite a lot of expertise in your
Engineering Department, but I don't have any names.

I know. I used to be one of them. More than twenty five years ago I
was writing code to get rid of aliasing artefacts in monochrome
256x256 digital camera images :-) That was back in the old days when
the only way to acquire a digital camera was to make it yourself by
unsoldering the metal top of a military spec dynamic RAM chip and then
focusing an image on it with a lens.



That's interesting. Do you know of current algorithms that do
something other than remove colour artifacts? I haven't seen any (but
I never looked carefully into the matter).

acl wrote:
On Oct 28, 1:26 pm, "David J Taylor"
[]
One problem with filtering in the optical domain may be that in the
computational domain, negative values of light intensity are allowed,
whereas they don't exist for broadband incoherent light detected by a
silicon sensor.

Are they possible for monochromatic and/or coherent light?

I don't know - that's outside my area of expertise, and I no longer have
contacts I could ask.

So in the optical domain you are constrained to a
positive impulse response for the filter (I think, corrections
welcome!).


One thing I never understood is that people keep talking about ideal
filters as step functions. But such a filter would cause ringing at
the cutoff frequency); in fact, there is always a tradeoff between how
much power you allow to pass in frequencies higher than your "cutoff"
and how much signal distortion you get back in real space. I've seen
several papers discussing this (mainly by introducing different
metrics and quantifying several different filter shapes).

So here's my question: you seem to know about such filters in audio.
What is done there? I know that oversampling is common, but do people
normally strive for sharp cutoffs? Is this tradeoff important there?

Yes, it's a trade-off. With 44KHz sampling you might have an analog
filter with roll-off from, say, 19KHz to minimise the overshoot. But with
oversampling at, say, 192KHz, the analog filter can start its roll of at
22KHz and be down to zero by 96KHz - easy! The rest of the filtering is
done in the digital domain where design and control of the filter response
is so much easier. I don't work in that field, but I believe that much
professional audio is now recorded at 96KHz or 192KHz, and only finally
down-sampled to 44.1KHz for CD production.

Cheers,
David

On Oct 28, 3:05 pm, "David J Taylor" -this-
bit.nor-this-bit.co.uk wrote:
acl wrote:
On Oct 28, 1:26 pm, "David J Taylor"
[]
One problem with filtering in the optical domain may be that in the
computational domain, negative values of light intensity are allowed,
whereas they don't exist for broadband incoherent light detected by a
silicon sensor.

Are they possible for monochromatic and/or coherent light?

I don't know - that's outside my area of expertise, and I no longer have
contacts I could ask.

I was using the Socratic method! Negative intensity is impossible for
both. But negative amplitude is possible (eg people do it when they
interfere laser beams), which I imagine is what you actually had in
mind when you made the qualification... I can't see how this could be
used to make anything useful as a filter though, so what Kennedy said
is probably going to stay true for some time.





So in the optical domain you are constrained to a
positive impulse response for the filter (I think, corrections
welcome!).

One thing I never understood is that people keep talking about ideal
filters as step functions. But such a filter would cause ringing at
the cutoff frequency); in fact, there is always a tradeoff between how
much power you allow to pass in frequencies higher than your "cutoff"
and how much signal distortion you get back in real space. I've seen
several papers discussing this (mainly by introducing different
metrics and quantifying several different filter shapes).

So here's my question: you seem to know about such filters in audio.
What is done there? I know that oversampling is common, but do people
normally strive for sharp cutoffs? Is this tradeoff important there?

Yes, it's a trade-off. With 44KHz sampling you might have an analog
filter with roll-off from, say, 19KHz to minimise the overshoot. But with
oversampling at, say, 192KHz, the analog filter can start its roll of at
22KHz and be down to zero by 96KHz - easy! The rest of the filtering is
done in the digital domain where design and control of the filter response
is so much easier. I don't work in that field, but I believe that much
professional audio is now recorded at 96KHz or 192KHz, and only finally
down-sampled to 44.1KHz for CD production.


Excellent, thanks for the clear reply.