Velocity Reviews - Computer Hardware Reviews

Velocity Reviews > Newsgroups > Computing > Digital Photography > Beyond Megapixels...

Reply
Thread Tools

Beyond Megapixels...

 
 
JPS@no.komm
Guest
Posts: n/a
 
      05-02-2004
In message <(E-Mail Removed)>,
http://www.velocityreviews.com/forums/(E-Mail Removed) (Dave Haynie) wrote:

>Of course it does. But George doesn't even begin to understand the
>issues he talks about, that much is clear. I doubt he could explain
>alising or imaging in digital sampling systems to save his life.


Does George's posting style look familiar to you? You knew him in
another life (newsgroup), under the name Steve Giovanella or "SG", in
comp.sys.amiga.advocacy.
--

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

 
Reply With Quote
 
 
 
 
George Preddy
Guest
Posts: n/a
 
      05-03-2004
(E-Mail Removed) (Dave Haynie) wrote in message news:<(E-Mail Removed)>...
> On Thu, 29 Apr 2004 18:50:48 +0200, "Bart van der Wolf"
> <(E-Mail Removed)> wrote:
>
> >
> >"George Preddy" <(E-Mail Removed)> wrote in message
> >news:(E-Mail Removed) om...
> >> "David J. Littleboy" <(E-Mail Removed)> wrote in message

> >news:<c6prm2$drg$(E-Mail Removed)>...
> >SNIP
> >> > But this is, as has been said before, incorrect: Foveon insists
> >> > that they don't need an antialiasing filter, which is simply a lie.
> >> > (The mathematics of sampling require that the signal be low-
> >> > pass filtered before sampling:

>
> >> A blur filter is needed with Bayer because of color sampling.

>
> >Wrong. ALL sampling needs low-pass fitering to reduce aliasing.

>
> Of course it does. But George doesn't even begin to understand the
> issues he talks about, that much is clear.


Foveon doesn't sample color, it measures color directly so a blur
filter is never needed.

Anytime you use sampling to predict an unmeasured outcome, like Bayer
does with all its output pixel colors, two things must first be true:
The measured attribute of the population must be randomly distributed
throughout, colors in an image obviously are not, and errors must
offest, which is also untrue for imaging. So interpolating a result
between two color samples can never hold any optical value whatsoever,
not under any circumstance.

Foveon directly measures full color and full luminance (like film).
1/4th samples are not used to interpolate (guess) color and luminance
from smaples like Bayer does, which then absolutely requires blur
filtering. Thus Foveon and film take "photographs," while Bayers
absolutely do not and can not.

Since you say you don't have a mathematical background, this analogy
is simple and exact: Bayer is like a poll that attempts to predict
the outcome of an election by sampling attributes from only a small
part of the polulation, while Foveon shows the election result itself.
One (Bayer), uses sampling to interpolatively guess all of the full
colors in an image (always incorrectly, since the colors in an image
are not randomly distributed), while the other (Foveon), is simply a
direct color measurement, so no blur filter is ever required.
 
Reply With Quote
 
 
 
 
George Preddy
Guest
Posts: n/a
 
      05-03-2004
Roland Karlsson <(E-Mail Removed)> wrote in message news:<Xns94DABF4443658klotjohan@130.133.1.4>...
> Fuzzy Logic <(E-Mail Removed)> wrote in
> news:Xns94D996C43D171bobarcabca@198.161.157.145:
>
> > My contention on this issue is that the average photographer does not
> > need more megapixels, they need better megapixels. I would much, much
> > rather own a 5 megapixel digicam with good resolution at high ISO
> > sensitivities and tough shooting situations than an 8 megapixel camera
> > with great resolution only in ideal situations.

>
> This is correct.
>
> I took some pictures in bright sunlight over the sea today. It would
> be just fantastic if I could catch the tones in the reflections in
> the water at the same time as the rest of the picture. More dynamic
> range is the most important quest right now!


This is why X3FL is currently a prerequisite to pro results. You
can't get DR like this without Foveon's ground breaking new
technology...

http://www.pbase.com/image/24653641/original

Bayer images with good shadow detail have white (blown) skies.

> > Cameras that
> > utilize Foveon's X3 sensor produce smaller images, but they are much
> > sharper, as red, blue and green color channels are captured in every
> > photosite, as opposed to the more standard use of Bayer interpolation.

>
> This sounds good in theory, but the Foveon implementation
> is lacking.
>
> They omit an anti alias filter.


No reason without color-sampling via a mosiac. Foveon measures full
color and luminance directly, not just samples.

> I know that som prefere
> the "sharper" pictures you then get. But ... at some stage
> in our evolving skill in looking at digital pictures most
> of us will see that something is wrong. The sharpness is
> faked. It might look good - just as a any artificial image
> might look good - but it is faked, and most of us will eventually
> be able to see that. All textured areas (grass, greenery, rock,
> water, skin, ...) has a very unnatural look. It is not all about
> fences and zebras - it is all over the picture. And - it cannot
> be removed - as aliasing produces artefacts that might be larger
> than the pixels.


Foveon images are not faked, they are real.

> Moreover - the filtering technique is not good. The color
> sensitivity of the three layers is weird. Converting leads to
> lots of noise and also faulty colors.


Foveon has less noise at ISO 800 than Canon has at ISO 100...

http://www.dpreview.com/reviews/sigmasd10/page13.asp

(Canon's ISO 800 is a disaster, my God look at that mess!)
 
Reply With Quote
 
Dave Haynie
Guest
Posts: n/a
 
      05-03-2004
On Thu, 29 Apr 2004 19:23:58 +0200, (E-Mail Removed) (Philip
Homburg) wrote:

>Anti-alias filters will reduce sharpness. For
>that reason, using a sensor a higher resolution than you would need in
>theory might be a good idea.


No. If you're filtering at the Nyquist frequency for a particular
sampling rate, you will NOT reduce sharpness at all in the final
resulut. Yes, you're reducing sharpness, but not in any way meaningful
to the sampling process -- you're removing spatial frequencies that
can't possibly be sampled, anyway.

If you're filtering above the Nyquist frequency, you won't eliminate
aliasing. That's going to be a bad idea, period.
Dave Haynie | Chief Toady, Frog Pond Media Consulting
(E-Mail Removed)| Take Back Freedom! Bush no more in 2004!
"Deathbed Vigil" now on DVD! See http://www.frogpondmedia.com
 
Reply With Quote
 
Bart van der Wolf
Guest
Posts: n/a
 
      05-03-2004

"George Preddy" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed) om...
SNIP
> Foveon doesn't sample color, it measures color directly so a blur
> filter is never needed.


You are sooo wrong, and you know it.

Bart

 
Reply With Quote
 
Bart van der Wolf
Guest
Posts: n/a
 
      05-03-2004

"George Preddy" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed) om...
> Roland Karlsson <(E-Mail Removed)> wrote in message

news:<Xns94DABF4443658klotjohan@130.133.1.4>...
SNIP
> > I took some pictures in bright sunlight over the sea today. It would
> > be just fantastic if I could catch the tones in the reflections in
> > the water at the same time as the rest of the picture. More dynamic
> > range is the most important quest right now!

>
> This is why X3FL is currently a prerequisite to pro results.


Wrong. Postprocessing cannot improve dynamic range, it only changes
tonality.

Roland is right, High Dynamic Range Imaging (HDRI) is the next big issue to
tackle. Lots of research has already been done, but the masses haven't
discovered it yet. Only a few professionals like ILM with their EXR file
format (http://www.openexr.com/), and a few other proprietary ones (), can
handle the huge dynamic range. An example of the possibilities, combined
with postprocessing to fit in the limited dynamic range of the output media,
can be seen here:
http://www.cs.huji.ac.il/%7Edanix/hdr/results.html
http://www.cs.sunysb.edu/~ash/tm/

And some attempts for dynamic Panoramas:
http://webuser.fh-furtwangen.de/~der.../hdrtest2.html
http://www.fieldofview.nl/rd-adr.php

Bart

 
Reply With Quote
 
Skip M
Guest
Posts: n/a
 
      05-03-2004
"George Preddy" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed) om...
> Roland Karlsson <(E-Mail Removed)> wrote in message

news:<Xns94DABF4443658klotjohan@130.133.1.4>...
> > Fuzzy Logic <(E-Mail Removed)> wrote in
> > news:Xns94D996C43D171bobarcabca@198.161.157.145:
> >
> > > My contention on this issue is that the average photographer does not
> > > need more megapixels, they need better megapixels. I would much, much
> > > rather own a 5 megapixel digicam with good resolution at high ISO
> > > sensitivities and tough shooting situations than an 8 megapixel camera
> > > with great resolution only in ideal situations.

> >
> > This is correct.
> >
> > I took some pictures in bright sunlight over the sea today. It would
> > be just fantastic if I could catch the tones in the reflections in
> > the water at the same time as the rest of the picture. More dynamic
> > range is the most important quest right now!

>
> This is why X3FL is currently a prerequisite to pro results. You
> can't get DR like this without Foveon's ground breaking new
> technology...
>
> http://www.pbase.com/image/24653641/original
>
> Bayer images with good shadow detail have white (blown) skies.
>
> > > Cameras that
> > > utilize Foveon's X3 sensor produce smaller images, but they are much
> > > sharper, as red, blue and green color channels are captured in every
> > > photosite, as opposed to the more standard use of Bayer interpolation.

> >
> > This sounds good in theory, but the Foveon implementation
> > is lacking.
> >
> > They omit an anti alias filter.

>
> No reason without color-sampling via a mosiac. Foveon measures full
> color and luminance directly, not just samples.
>
> > I know that som prefere
> > the "sharper" pictures you then get. But ... at some stage
> > in our evolving skill in looking at digital pictures most
> > of us will see that something is wrong. The sharpness is
> > faked. It might look good - just as a any artificial image
> > might look good - but it is faked, and most of us will eventually
> > be able to see that. All textured areas (grass, greenery, rock,
> > water, skin, ...) has a very unnatural look. It is not all about
> > fences and zebras - it is all over the picture. And - it cannot
> > be removed - as aliasing produces artefacts that might be larger
> > than the pixels.

>
> Foveon images are not faked, they are real.
>
> > Moreover - the filtering technique is not good. The color
> > sensitivity of the three layers is weird. Converting leads to
> > lots of noise and also faulty colors.

>
> Foveon has less noise at ISO 800 than Canon has at ISO 100...
>
> http://www.dpreview.com/reviews/sigmasd10/page13.asp
>
> (Canon's ISO 800 is a disaster, my God look at that mess!)


There's something truly odd about that image. The horizon is jagged, and
the side of the rock that is in shadow has a dark edge, like it was cut and
pasted in. Is this the best that Sigma can do, George?
As far as your comparison at dpreview, look at the lousy colors that the
Sigma reproduces, and they get worse as the ISO increases, or as dpreview
says, "However, it's not all rosy for the SD10, what the crops above don't
show is what happens to the color patches as ISO increases. The EOS 10D
manages to maintain both the hue and saturation of the distinct red, green
and blue patches all the way through to ISO 1600, at ISO 3200 all the
patches are affected by the high noise level, noticably the blue patch is
turning purple. The SD10 gradually loses both saturation and hue accuracy as
sensitivity is increased, up to ISO 400 things appear to be acceptable, at
ISO 800 it's borderline with the red patch taking quite a lot of yellow /
green, at ISO 1600 noise levels in the color patches is quite strong."
Further, I wouldn't say the colors were acceptable, just that the
consistency of the Sigma colors was acceptable. But that's my opinion, not
that of dpreview.

--
Skip Middleton
http://www.shadowcatcherimagery.com


 
Reply With Quote
 
Dave Martindale
Guest
Posts: n/a
 
      05-03-2004
>On Thu, 29 Apr 2004 19:23:58 +0200, (E-Mail Removed) (Philip
>Homburg) wrote:
>>Anti-alias filters will reduce sharpness. For
>>that reason, using a sensor a higher resolution than you would need in
>>theory might be a good idea.


(E-Mail Removed) (Dave Haynie) writes:
>No. If you're filtering at the Nyquist frequency for a particular
>sampling rate, you will NOT reduce sharpness at all in the final
>resulut. Yes, you're reducing sharpness, but not in any way meaningful
>to the sampling process -- you're removing spatial frequencies that
>can't possibly be sampled, anyway.


Actually, Philip has a point. An anti-alias filter that removes all
modulation at the Nyquist frequency will also reduce the amplitude of
frequencies somewhat below the Nyquist frequency. These frequencies can
be sampled correctly; the problem is that a "brick wall" filter that has
no effect below Nyquist but complete attenuation above Nyquist is not
physically possible. This does reduce apparent sharpness below what is
possible for a given pixel count. Some of the lost sharpness can be
restored by suitable sharpening, but that also increases noise.

Oversampling is one solution to this, if it's practical. For 2X
oversampling, you'd shoot the image at double the pixel density (4X as
many pixels) as you ultimately need, then resample down to the desired
output size. Since the resampling is done digitally, the low-pass
filter can have as sharp a cutoff as you're willing to pay for in
computer time. This is how digital audio is usually captured, although
the oversampling factor is usually a lot larger than 2X.

Dave
 
Reply With Quote
 
Roland Karlsson
Guest
Posts: n/a
 
      05-03-2004
(E-Mail Removed) (George Preddy) wrote in
news:(E-Mail Removed) om:

> Foveon doesn't sample color, it measures color directly so a blur
> filter is never needed.


he he . You clown


/Roland
 
Reply With Quote
 
Roland Karlsson
Guest
Posts: n/a
 
      05-03-2004
"Skip M" <(E-Mail Removed)> wrote in
news:vbqlc.12948$ph.4919@fed1read07:

> There's something truly odd about that image. The horizon is jagged,
> and the side of the rock that is in shadow has a dark edge, like it
> was cut and pasted in. Is this the best that Sigma can do, George?


At the first glance it looks like an impresive image, although
I have never ever seen a sky this kind of blue.

But when you look at it closer you see the strange horizon -
lots of aliasing there. Looking even further you see lots
of artefacts everywhere. All surfaces, including the water,
looks weird. To really see whats going on you have to (at least
on my monitor) enlarge the picture to 200%. Now you see that
all edges has strange banding artefacts and that all textures
looks like tetris bricks. It is not all that easily visible
in this image. This is probably because the lens is not all
that much sharper than the detector, or maybe it is not 100%
correct focussed. With a sharper lens, the picture would be a
disaster.

And this is the main objection to no anti alias filter - the
sharper picture, the more artefacts.


/Roland
 
Reply With Quote
 
 
 
Reply

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
Wap and beyond Frank Wireless Networking 1 06-18-2005 03:00 AM
How to get beyond a shared internet connection =?Utf-8?B?bWFyYnk=?= Wireless Networking 0 01-04-2005 03:33 AM
Re: Why is non-routeable 10.100.10.x routing far beyond my network? Peter Cisco 0 11-15-2004 08:31 AM
Any way to increase logging level beyond debugging? Forrest Cisco 0 03-05-2004 10:05 PM
VHDL: practical questions: beyond just hobbies Neil Zanella VHDL 0 11-16-2003 06:46 AM



Advertisments