purple fringe - due solely to Bayer filter?

OK, here's my theory (no doubt 10,000 flames inbound ).

We all know that the Bayer filter is short of data in the red and blue
channels actually they have only half the data of the green channel.

So imagine an image where an area of 100% white is right next to an
area that is substantially less.

The de-mosaic / interpolation process doesn't know where the dividing
line is and so interpolates ie invents an intermediate value for each
channel. Because the red / blue channels have 1/2 the data this is
more obvious in red / blue ie you get a purple fringe.

I have looked at loads of test images of on dpreview, stevesdigicams
etc and all Bayer cams have the fringe.

Please note that this is not a Bayer / Foveon post, just my thoughts
on this one problem.

Cheers!

HB

 

Wow, I wish I had known this 30 years ago when I was blaming a cheap lens on
my film camera for the problem. How does your theory fit in with fringing on
Sigmas and film? As I recall, some have speculated that maybe the Sigma
cameras suffer the problem worse than some of the bayer cameras because they
capture the image more precisely.

Well, you asked for thoughts, I personally think this theory needs a little
work

For what it's worth

Jeff

 

Nothing to do with that. Chromatic aberration (aka. purple fringing) is
due to the lens focusing red and blue light in slightly differently
places. This is what so-called achromatic lenses in more expensive
lenses are designed to solve. CA can also arise due to the microlenses
in front of the sensor - both Bayer sensors and the SD10's Foveon sensor
use these. Foveon and Bayer sensors are equally prone to this effect.
Indeed, problems with chromatic aberration is a consistent (minor)
criticism of the Sigma SD10 in some reviews. The SD9 does not suffer so
much from this - other than due to the camera's lens - as it has no
micro lenses over the sensor. But as a consequence, its low light
performance is abysmal and it can go no faster than ISO400.

 

I think you can very crudely determine whether your theory is correct.
Shoot an image of a checkerboard. As the sensor filter is the same all
the way across, you'd expect the same effect (on the same side of the
light/dark transitions) at both ends, if it were the source of the
problem. However, if the fringing is to the left on one side, and to the
right on the other, you can reasonably assume that the problem is
optical. It's either the lens itself, or the angle of incidence on the
sensor itself.

 

The effect you are talking about - highlight-created purple fringes -
seems to be Bayer specific and should not be confused with magenta and
green, or yellow and blue, chromatic aberration fringes. The Sigma picks
these up as badly as any camera, and they can be correct by the same
chromatic aberration controls in things like Photoshop's CS Raw import
plug in. The Sigma does not have the 'purple fringe' phenomenon at all,
but then nor do most larger sensor DSLRs. It's mainly found in small
sensors and especially in 2/3rds 8 megapixel sensors.

Makers are still trying to eliminate it but the safest way is to use a
small aperture. It seems to have something to do with lens aberrations
combining with the way microlenses work on a Bayer CCD. At the least it
may just be a non-chromatic aberration - coma - making itself seen as it
used to on old photos.

David

 

(digiboy) wrote in
Background:
The normal purple fringing problem is radial symmetric and
is therefore either due to lens color aberration or due to the
incident angle the light hits the sensor. The microlenses are
suspected to be one cause in the latter case.

If this is true, then the fringe shall be no more than 1 pixel wide.
It shall also be striped in blue/red along the edge. Now - all good
interpolation algorithms I have seen smooths the chroma information.
So - the edges should (in a perfect world) be the same color as the
dark area in your example.

Can you please point at some typical examples?


/Roland

 

Gee another Geroge Preddy alias.

 

Please bear in mind there are two quite different types of chromatic
aberration.

Longitudinal or axial chromatic aberration (LCA) is the one you have
described, i.e. the effect of a lens bringing light of different colour
to a focus at different distances. It is improved by stopping down, and
is constant across the field. This is not what is being observed here.

The other, which is the one which normally causes colour fringing, is
transverse or lateral chromatic aberration (TCA), which is a difference
in magnification of the image from light of differing wavelengths. TCA
is an off-axis aberration; it gets worse at greater angles of incidence,
i.e. towards the edge of the field. It is not at all improved by
stopping down. The fringes are normally blue toward the axis and red
toward the edge of the image (though I suspect over-correction is
possible, I doubt it occurs often). Interestingly, it gets worse for
longer focal lengths, and is why fluorite or very expensive special
glasses need to be used to make such lenses perform well.

The effects you mention, if they arise from the taking lens, will almost
certainly be TCA, which is visible in most lenses if you look close
enough. Strangely enough, it is one of the easiest to correct using
software. Bryan Caldwell, whose optical expertise is orders of magnitude
ahead of mine, has explained this several times here before, and has a
very interesting web site on how to do it.

You should be able to check if TCA in the taking lens is causing the
problem; if the fringes get worse at the edge, and re absent or very
small at the centre, then it is very likely* that TCA is the cause. If
it is constant across the frame it certainly is not.

*I say "very likely" rather than "certain" as, of the two other causes
discussed, sensor angle of incidence also increase towards the edge.

Other tests: If the fringes also occur on a film body and get worse at
the edge, it's certainly TCA.

If the fringes get worse with a telephoto lens, it's almost certainly
TCA.

If the fringes get less with a long lens and worse with a wide, it's
probably a sensor incidence angle/microlens issue and has nothing to do
with the taking lens.

If the fringes stay pretty well the same regardless of lens and position
in the frame, it could well be a sensor artefact. How this could be
proved will require expertise beyond anything I can offer.

 

Sigh. Where do you get this?

Look at this photo:
http://www.pbase.com/image/27849780

Look at the man in the white t-shirt near the center of the frame with
the sunglasses on his head. Look at the specular reflection of the sun
on the sunglasses. See the purple fringe? (It's not the lens: the
comments say that this was taken with the 50mm EX, arguably Sigma's
best lens.)

Early versions of the SD9 were notorious for this on many specular
highlights. The SD10 seems to be better, but it still happens.
It's also not uncommon on specular highlights on Kodak SLR/n photos.
It's most likely a blooming artifact (charge leakage to adjacent cells.)
The easiest way to reproduce it is to take a photo of a shiny black
car (even better if it has lot's of chrome) in the sun.

It does, it just takes a brighter light/dark boundary to see it.

(BTW, if it's only a "small sensor" artifact, how can you say it's
Bayer specific when we don't have any images from a small sensor X3
camera?)

 

This isn't the same purple fringe as people are talking about. I am
aware the Sigma can bleed over when shooting a light source - I've done
it. Interestingly, the Sigma produced an identical effect to the
halation exposure visible on Fuji Velvia from the same shot with the
same lens, which leads me to think that the sensor may just be copying
an effect also visible on film.

If you've seen the purple fringes which people complain about from the
Sony, Canon and Minolta 8 megapixel cameras; from the Minolta Z1/Z2 and
its Kodak brothers; and one or to other cameras - they are directional,
affecting one side of a highlight area typically, and not chromatic
(i.e. they do not have any matching complementary coloured fringe on the
opposite transition of light/dark in the same zone of the image).

The Sigma doesn't produce this kind of fringe, and nor do most DSLRs.
The new Kodak DCS Pro/n produces something similar but not identical. I
have yet to try one with the firmware fix. In my opinion, the Pro/n
suffered from an entirely different problem with chromatic shifts in the
peripheral field emphasised on one side by a tendency to a blue
chromatic aberration-like error.

David

 

Whether it's the same artifact (or a related one) is a matter for
speculation. It depends on if you attribute it to blooming or
lens/microlens interaction or just lens. But nothing in your
"directional purple" explanation suggests one way or another how the
color filter array system contributes to the artifact. Indeed, if it
were a mosaic induced problem, you might expect the color shifts to be
different for different parts of the image because the filter array is
not perfectly symmetric.

Or it could be lens related. See http://www.pbase.com/image/24586310

OK, then like I said, you cannot conclude anything about X3 vs. CFA
from this evidence.

 

Too bad you don't have any knowledge of the subject. You might have been
able to express a valid opinion if you had done some study.

--
http://www.chapelhillnoir.com
home of The Camera-ist's Manifesto
The Improved Links Pages are at
http://www.chapelhillnoir.com/links/mlinks00.html
A sample chapter from my novel "Haight-Ashbury" is at
http://www.chapelhillnoir.com/writ/hait/hatitl.html

 

I note the "full exif" data is empty...

 

It's not the CAMERA lens, but it could be explained away by the
SUNGLASSES lens.

A better place to see the chromatic aberrations is at the top of image.
The chrome on the cars is doing a splendid job of making purple.

In any case, purple fringing has been around long before Bayer or Foveon
sensors. It's symptom of poor lens design.

This image SHOULD be more problematic, yet it isn't:

http://www.pbase.com/image/26193750/original.jpg

 

Like I said, that shot of Champa's used the Sigma 50mm EX macro @ f8
AND it's near the center of the frame. It should not be the lens. It's neither
directional nor the same color as in http://www.pbase.com/image/24586310.

You can also find similar examples of blooming in the Sigma sample
portrait in the highlights of the eyes (but you have to blow up the
image quite a bit tosee them.)

I'm not sure why you think this photo SHOULD be problematic. It has
none of the characteristics where you most commonly find the problem:
* no dark outline (of foliage) against a bright sky
* no specular highlights against a dark background.

 

well I was right to suppose I would get a load of flames

Lets be clear. I have no axe to grind for or against Bayer. I have no
axe to grind for or against Foveon.

All I was saying is that if an image has a very high contrast
transition occuring spatially in a small area, and you have to
interpolate a channel to try to make up for another channel having
more data, the interpolated channel will be incorrect, yes?

Surely nothing too controversial. There are bound to be other effects
at work.

But all I can say is that if you get the high res pictures off
DPReview and enlarge them in say photoshop then you will see whether
you get a fringe.

So I did.

And this is what I found:

The DC14N has fringing. The DC14N doesn't have microlenses
The D100 has fringing. the D100 does have micro lenses
The Sigma doesn't have fringing. the Sigma doesn't have microlenses

OK I could go on. If microlenses give you a fringe, why does the DC14N
have a fringe?

As I have no axe to grind in any direction, the only logical
conclusion I was able to come to was that the fringing is due to the
method of imaging.

 

I had a look at this image, and zoomed right in using PS. One of the
pixels you refer to measures as R251 G5 B234 so is basically magenta.

This isn't really the cirumstance I am referrring to.

If a Bayer had taken this image the mans white T shirt would have a
fringe on his right sleave, like in this image from a D70
http://img2.dpreview.com/gallery/nikond70_samples/originals/dsc_0063.jpg.
Have a look at high contrast transition areas.

As I said before I have no axe to grind.

 

your thoughts are fair enough. I was always lead to believe that
chromatic aberation caused the different wavelengths of lights to
focus at a different point. If so a lens with chromatic aberation
might give a Foveon serious problems?

fwiw I have never seen a coloured fringe on any of my films. I can't
claim 30 years worth of films, but I can claim 20 years

 

yes, and its the fact that on my 2 Canons the fringe is always to the
right that set me thinking. Sometimes the fringe is on the right and
lower in the frame.

But alot depends on the demosaic algorythm. If the sensor is short on
data horizontally it'll be the same vertically as well. So this
fringing is no surprise to me.

I've no problem with lens problems interferring (so to speak) as well.

 

If microlenses are the problem why does the Kodak DSC14 get fringing?