[Leica] aliasing

[jean.louchet at inria.fr (Jean Louchet)]

Hi, it's me again :-)

> 
> From: Adam Bridge <abridge@gmail.com>
> Subject: Re: [Leica] Fungus in camera
> To: Leica Users Group <lug@leica-users.org>
> Message-ID: <4cfa589b040701233643a4a3ac@mail.gmail.com>
> Content-Type: text/plain; charset=ISO-8859-1
> 
> It seems very odd to me that Leica is trying to eliminate the
> anit-aliasing filter. I distinctly remember my digital signal
> processing courses that state that before you sample an analog signal
> you have to limit the bandwidth with a filter to at least half the
> sampling frequence (and that would be if you had a perfect low-pass
> filter, you have to go lower for real-world filters.)

Yes, _before you sample_, not after. The point is that the word "filter"  
has 2 different meanings. The only filter that will work BEFORE sampling
is an optical filter (a piece of glass). After sampling, the only possible
resource is digital filters (image processing algorithms) but the harm has
already been done, and you just can't recover information that has been
lost (as I wrote in a recent posting).

> Moire patterns would be the result because artifacts would be generated
> by digital sampling.

YES but how could the algorithm make the distinction between an artefact
moire and the pattern on one's shirt?

> I had a real-world example of this where we were
> measuring ion-acoustic waves in very low-density argon plasmas and got
> results that were really exciting - until the experimentalist realized
> he hadn't built-in the low-pass filter. When the filter was installed
> the "exciting" results vanished and things approached theory. Oh darn.
> But that wasn't with optical systems and maybe someone who's won the
> galactic institude prize for extreme cleverness has figured out how to
> avoid the artifacts.
> 
I suppose you used an analog filter BEFORE sampling to solve the problem 
(in acoustics, analog filters are not pieces of glass but probably some 
mechanical device?). Once sampled it's too late!

> From: Peter Dzwig <pdzwig@summaventures.com>
> Subject: Re: [Leica] Fungus now filters
> To: Leica Users Group <lug@leica-users.org>
> Message-ID: <40E559B5.6060404@summaventures.com>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>
> The point of anti-aliasing is to smooth out "the jaggies", the jagged
> edges in digitised images caused by finite resolution.

You are referring to what Computer Graphics people call "antialiasing"  
but the same word does not refer to the same reality. What they call
antialiasing is the set of rendering techniques to smooth out edges (like
the Bresenham algorithm). The aliasing problem found in digital cameras is
another issue, it is a sampling problem as Adam said, not a rendering
problem.

> In theory with a large enough number of pixels in a camera you should be
> able to ignore the effects because the human eye wouldn't be able to
> resolve them.

This is not true. Even with as many pixels as you want, if the lens
resolution is good enough to transmit spatial frequencies higher than half
the pixel frequency AND the sensor elements are not contiguous, there will
be aliasing which will generate low- and medium-frequency interferences
that anybody will be able to see (even without a very good sight).

> The need for anti-aliasing is a result of the artifacts introduced by
> the availability of a finite number of pixels to display an image in.
> For a standard monitor this number is 1.25Mp and here anti-aliasing is
> necessary without doubt.

No, you are again mixing up two different problems, one is sampling and 
the other one is rendering.
> 
> It MAY be - and I say may because I don't know for certain - that at 14
> MP, up to certain "reasonable" magnifications the effects are either not
> noticeable or are swamped by other effects or are cancelled out by the
> effects of other algorithms, when the digital image is viewed or 
> printed.

I don't agree. Of course aliasing will be reduced but this is just because
lenses act as lowpass filters. With high quality lenses there will still
be aliasing with 60MP. As I wrote last week, the two only methods are: not
using a lens "better than the sensor", or designing a sensor where there
is no gap between adjacent photosensitive elements (this will also act as
an analog lowpass filter). The computer graphist's antialiasing is just
digital (algorithmic)  antialiasing and can't do anything to repair a
violation of Nyquist-Shannon's rule.

> On Thu, 01 Jul 2004 08:49:02 -0400, Dan C <bladman99@yahoo.ca> wrote:
>> The other article concerns the digital back for the R8/R9.  They are
>> aiming for a December launch, but they seem to be having problems
>> getting the internal image processing software ready.  Leica isn't
>> happy with it.

My own guess is that Leica people (who are great optical and mechanical
engineers but maybe not digital image processing specialists) are
discovering that digital image processing is hopeless if sampling was not
good enough. R lenses are so sharp that aliasing problems must be really
hard and digital sensors not up to the task. Again, non-adjacent
photosensitive elements are a real problem with good optics. There is no
aliasing trouble in the Espio at 400x640 pixels with a cheap plastic
single element lens!

Jean

 ------------------------------------------------------------
 Dr Jean Louchet       COMPLEX Project     INRIA Rocquencourt
                       BP105   78153 Le Chesnay cedex, France
 Jean.Louchet@inria.fr     http://fractales.inria.fr/~louchet
 ------------------------------------------------------------