Velocity Reviews - Computer Hardware Reviews

Velocity Reviews > Newsgroups > Computing > Digital Photography > Anti-aliasing, Color Separation, etc.

Thread Tools

Anti-aliasing, Color Separation, etc.

Rodney Myrvaagnes
Posts: n/a
On Fri, 19 Dec 2003 19:30:11 GMT, "Paul H." <(E-Mail Removed)>

>"Rodney Myrvaagnes" <(E-Mail Removed)> wrote in message
>news:(E-Mail Removed).. .
>> On Fri, 19 Dec 2003 06:36:12 GMT, "Paul H." <(E-Mail Removed)>
>> wrote:

>> The site is so organized that a beginner wouldn't be able to do that,
>> as apparently you also are not.

><Sigh> Now, now...

I said "apparently" because you didn't.

>> I took it at face value until I
>> noticed the grossly impossible color-response graph for stacked
>> sensors and the absence of a "Disadvantages" list for the stacked
>> sensor.

>Grossly impossible? Grossly over-simplified, perhaps, but no more so than
>was the related graph for the Bayer-type sensor; I simply thought the graphs
>were intended to be schematic and illustrative rather than truly
>mathematical in nature.

It was illustrative of a fantasy, and contradicted the author's claim
about the stability of silicon absorption making the stacked system

If he hadn't made the claim about silicon absorption, then the graph
could have been taken as only grossly oversimplified. But, since he
tied the stacked sensor to silicon absorption, he was obliged to keep
the good with the bad for that system. He did not.

This is the kind of stuff that editors of engineering magazines have
to be alert to all the time. IEEE Spectrum has enough prestige that it
can force authors to submit to peer review by their competitors'
marketing managers.

Others, including the one I worked for, were not so fortunate. We had
to keep after people constantly or risk losing credibility with the
readers. Without readers, no advertisers.

> I must admit, though, I had overlooked the lack of
>a "disadvantage" list for stacked sensors and that is a real oversight by
>the author. However, after re-reading the page I still don't see the
>extreme bias to which others have referred:

If you had worked as an editor you would have. Did you notice the word
"plastic" inserted in the Bayer sensor description? A loaded word
implying cheapness, and in now way an automatic part of a Bayer
sensor. Contrast the stable silicon absorption as an included part of
stacked sensors.

> Indeed, even the conclusion
>comparing stacked sensors to Bayer-type sensors was fairly bland; at no
>point did the author ever say anything approaching "therefore, stacked
>sensors are far superior." Thus while "bias" may be present, I don't see
>real advocacy and I still think the phenomenon of spatial aliasing was
>illustrated effectively.

>What I don't understand, however, is the extreme dislike of Foveon/Sigma
>exhibited by many on this newsgroup: after all, Foveon sensors represent a
>mere blip in the digital camera marketplace and, to me, at least seem more
>of an interesting technical curiosity than a threat to the future of digital

You are right that it poses no threat to anything. If it had any value
it would be an advance, not a threat, to digital cameras in general.

My impression, from reading this NG, is that the dislike is not for
Foveon/Sigma per se, but for dishonest marketing. I expect any of the
detractors would buy a Sigma lens that did something they needed at a
better price than Nikon or Canon.

But, I also expect they woud be wary of any Sigma camera after Sigma
attempted to foist off DSLRs lacking an AA filter. There is no way to
fix that after the fact.

>However, what I was trying to do was NOT contribute to the
>controversy, so if I can find some other elementary and far less
>controversial sites, I'll post the url's, then put my shields up.

Rodney Myrvaagnes J36 Gjo/a

"In this house we _obey_ the laws of thermodynamics." --Homer Simpson
Reply With Quote
Posts: n/a
In message <(E-Mail Removed)>,
Rodney Myrvaagnes <(E-Mail Removed)> wrote:

>But, I also expect they woud be wary of any Sigma camera after Sigma
>attempted to foist off DSLRs lacking an AA filter. There is no way to
>fix that after the fact.

That's the SD10. The SD9 doesn't even have microlenses, so only 30% of
the light hitting the sensor hits a photosite. This reduces sensitivity
by a stop and a half, making it even harder to interpolate the foveon
color. It also guarantees aliasing at a level where any sharply focused
image is not a continuous image at all, but a collection of pixels such
as you would have if you had a much higher resolution, properly filtered
image, and only took the rows and columns of pixels that were a multiple
of 3 or 4.


<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <(E-Mail Removed)>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><

Reply With Quote
Posts: n/a
"David J. Littleboy" <(E-Mail Removed)> wrote in message news:<brut7p$3ev$(E-Mail Removed)>...
> >
> > In a Bayer sensor without AA filter,

> There's never a Bayer sensor without an AA filter. (Except for Kodak 14n and
> some of the MF digital backs, and the 14n exhibits nasty artifacts.)

Correct, but it was a working hypothesis

> > In the Foveon sensor without AA filter, the luminance and chrominance
> > components are slightly aliased.

> There's also never a Foveon sensor without an AA filter. At least not on my
> shelf there isn't.

AFAIK, the SD9/SD10 don't have an AA filter in front of the sensor.

> The Foveon AA filter must completely, 100%, totally, attenuate all
> frequencies at and above Nyquist.

It should, in theory.

> Any practical AA filter that does that will, I suspect, be adequate for a
> Bayer sensor.

No. Particularly, the red and blue components would require a stronger
AA filter to be completely dealiased.

> If not, it will be very close to adequate.

Hum... "very close" is not very scientific, compared to your previous
"completely, 100%, totally"

> > Saying that the Foveon does not require AA-filter is false, or at
> > least a strong shortcut. But what is true is that the Foveon requires
> > less AA-filtering than Bayer's

> That _may_ be true. (See my other note.*)

That IS true.

> Seriously investigating what
> the practical effects of that difference would be is an interesting
> question. But that's not what Foveon is doing, so it's irrelevant to the
> cameras we have as options. The better Bayer cameras are quite good at
> minimizing artifacts, so as a practical issue were faced with fairly
> reasonable camera implementations on the one hand and an incorrect
> implementation on the other.

Understand me. At the moment, if I wanted to buy a DSLR, I very
probably wouldn't buy a SD9/SD10. Both the sensor, the body, and the
post-processing of the raw data are not mature enough, I think.
However, what I want to point out is that from the scientific point of
view, the Foveon approach is very consistent.

> FWIW, if the SD9 and SD10 cameras had anti-aliasing filters, and if Foveon
> was arguing that Bayer cameras required a stronger AA filter and that the
> Foveon sensor could deliver, say 66% of the Nyquist frequency whereas Bayer
> cameras could only deliver 55% without objectionable artifacts, and that
> Foveon therefore delivered 66/55 or 20% better resolution than (or 1.44
> times as many megapixles as) Bayer sensors, I'd be on the Foveon side of the
> fence. But that's not what they're saying. (These numbers are guesses. It's
> probably more like 66% vs. 60%.)
> <<<<<<<<

Let's take sensors of 16*24mm (I konw that the Foeveon is actually
smaller, but for direct comparison it's easier to consider 2 sensors
with the same size). A 6Mpix Bayer has 2000x3000 photosites, the
3.4Mpix Foveon approximately 1500x2250.

Distance between photosites: 24/2250=0.0106mm
Nyquist frequency: 0.5/0.0106=47cycles/mm

Distance between photosites: 24/3000=0.0080mm
Distance between R/B photosites: 2*0.0080=0.0160mm
Nyquist frequency for R/B: 0.5/0.0160=31cycles/mm
(diagonal) distance between G photosites:
Nyquist frequency for G: 0.5/0.0113=44cycles/mm

If both sensors were equipped with the appropriate AA filter to remove
*all* the aliased frequencies, the final resolution of the Foveon
would be 47/31=1.52 times higher than the Bayer's one.

In practice:
-- Foveon doesn't use any AA: The R,G,B components above 47cycles/mm
are aliased.
-- It seems that a usual choice for the Bayers is to set up the
AA-filter so as to deliase the G component. The cut-off would be at
44cycles/mm. If the roll off were sharp, the resolution would then be
similar to the Foveon's (actually I guess that the roll off is not so
sharp and that the cutoff freq. is higher). The R/B components are
aliased between 31 and 44 cycles/mm.

Now, there's finally the low pass filtering effect of the photosite
size to take into account. Let's assume that the photosites size is
equal to the distance between them (no free space). The cut-off
frequency is 1 cycle/photosite size:
-- Foveon: 1/0.0106 = 94cycles/mm
-- Bayer: 1/0.0080 = 125cycles/mm
At half this frequency, the attenuation is 0.64, and it is only 0.90
at the quarter.

Taking into account this "natural" low-pass, we end up with:

-- Foveon:
-attenuation = 0.64 at 47cycles/mm (Nyquist)
-full attenuation at 94cycles/mm (2*Nyquist)

-- Bayer.
-attenuation = 1*0.90=0.90 at 31 Hz (B and R Nyquist)
-full attenuation at 44 Hz (G Nyquist)

The conclusions:
-- the two sensors (Foveon without AA Bayer with a usual AA) have
similar resolutions.
-- With the Foveon, all the components are aliased to some extent.
With the Bayer, the G component is unaliased, but the R/B components
are more aliased.

ne pas répondre à cette adresse / do not reply-to this adress
Reply With Quote

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are Off

Similar Threads
Thread Thread Starter Forum Replies Last Post
Changing font color from current font color to black color Kamaljeet Saini Ruby 0 02-13-2009 04:58 PM
standard way of getting color # string from drawing.color? PJ6 ASP .Net 4 09-15-2005 06:10 PM
Color.white vs. Color.WHITE Niels Dybdahl Java 3 10-06-2004 03:21 PM
how to convert color in style "#e6e6e6" to color object in moondaddy ASP .Net 3 04-28-2004 04:59 AM
BGR Color to Java Color Andrew Arace Java 7 09-16-2003 05:30 PM