Sigma/Foveon Questions

In article , Martin Brown
wrote:

Luminance is inferred from the Bayer data based on the GR,BG matrix or
in some cases CM,YG matrix. The luminance is dominated by the green
channel and the red and blue provide minor corrections to it through
various cunning heuristics.

that's not how it works. for any given pixel, it looks to at least the
surrounding pixels (3x3) and maybe two levels out (5x5). beyond that
isn't generally worth it. there is also no single algorithm either. an
awful lot of pixels contribute to rgb of one pixel.

Having the sharp edges mangled by subsampling faults stick out like a
sore thumb in the handful of cases where it is relevant.

not when it's chroma being subsampled. try it in photoshop. convert the
image to lab and blur the ab channels. you only need to blur it with a
2 pixel radius to simulate bayer, but you can crank it to 5 pixels and
not notice a difference.

That depends.

on what? i guarantee that you can't see the difference.

The red channel as reconstructed by standard JPEG decoders
can corrupt the luminance value by enough to be a nuisance in some
cases. It is only obvious when this situation arises which is typically
most obvious with fine black lines on near saturated red. This quirk is
part of the reason why JPEG images drift when recompressed many times.

this isn't about jpeg decoding though.

I am no great fan of the Foveon sensor, but it does have an edge for
certain photography and might be what the OP is looking for. However it
also has its own problems so only he can decide if it is for him.

that's true. unfortunately the types of photography for which it is
suited are very limited.

Bubba wrote:
On Apr 20, 5:04?am, Martin Brown
wrote:

The problem arises later in the imaging chain. Bayer sensor struggles a
bit with a pure red (or pure blue) monochrome images because it has
fewer independent pixels. Normally the luminance channel is able to hide
these defects, but when the situation arises where the luminance channel
is corrupted by the chroma channels then you lose detail.

Okay, now this interests me. I asked on other threads the obvious
question why a CCD sensor should be considered the equivalent of a
CMOS, if no very low-end P&S has a CMOS.

CMOS has an extra on-chip wiring overhead which matters a lot in very
tiny dense sensors, and not much on big DSLR sensors. The new
back-illuminated CMOS sensors remove that CMOS overhead (to the dark
side of the chip) and are bringing CMOS sensors into low-end P&S and
phone cameras.

Why would green not be a problem?

You didn't know that Bayer sensors have twice as many green pixels?

--
Chris Malcolm

In article
,
Bubba wrote:

CMOS has an extra on-chip wiring overhead which matters a lot in very
tiny dense sensors, and not much on big DSLR sensors. The new
back-illuminated CMOS sensors remove that CMOS overhead (to the dark
side of the chip) and are bringing CMOS sensors into low-end P&S and
phone cameras.

Does this mean (what someone reviewing a Canon SX1 said) that the CMOS
sensors on $400--$500 P&S are "small" and (I suppose) mediocre? How
can you tell which CMOS sensor a particular camera has, and whether
it's any good?

dpreview normally lists the size of the sensor. you can also determine
it by comparing the actual focal length of the lens versus the 35mm
equivalent focal length. the higher the ratio, the smaller the sensor.

for example, the canon sx1 has a 1/2.3" sensor, which is much smaller
than what's in an slr. the lens is a 28-560 equivalent but if you look
on the lens itself, it's actually a 5-100mm lens, making the conversion
factor 5.6. a typical slr is 1.5-1.6x, and you'd need an 18-375mm lens
for the same equivalent.

in other words, the sensor is tiny.

http://www.dpreview.com/reviews/canonsx1is/

You didn't know that Bayer sensors have twice as many green pixels?

No. That's why I post questions here. Why in God's name would they
have twice as many?

because there are three primary colours that need to fit into a 2x2
grid. there are a number of ways to arrange them, but doubling green
produces the best results since that's where the eye is most sensitive.

sony tried emerald as a fourth colour and kodak has a patent on using
no filter (i.e., white) but i don't think that's in a camera yet. some
cameras even used cyan, magenta, yellow and green.

I haven't bought a new digital camera in three
years, because of something I read here back then about my particular
problem--red channel flare--not ever improving in digital photography
unless (three years ago) you could afford a splendiferously expensive
camera ($+++Ks).

do you have an example of this so called flare?

In article , nospam
writes
In article
,
Bubba wrote:

You didn't know that Bayer sensors have twice as many green pixels?

No. That's why I post questions here. Why in God's name would they
have twice as many?

because there are three primary colours that need to fit into a 2x2
grid.

Nothing of the sort. There is no requirement for a 2x2 grid, the
colours could be arranged as triads, as in the dots on a conventional
colour CRT, or as linear triplets as in the lines of a Trinitron screen
or LCD & Plasma displays. Since each pixel has only one true and two
interpolated colours, the actual format of the colour filter array
doesn't make any difference to resolution.

The reason Bruce Bayer proposed the GRGB 2x2 matrix with twice the
number of green pixels as red or blue was to crudely approximate the
spectral response of the human eye, which is at least 3x higher to green
than red and 6x higher than to blue.
http://en.wikipedia.org/wiki/File:Eyesensitivity.png
It isn't close to an ideal match to the human eye response, but it is 2x
better than any equal triad solution and hence requires less matrix
manipulation, a major source of noise, to achieve correct spectral
response.

Mismatched spectral response is one of the reasons why the Foveon
concept isn't as good as its supporters claim. Whilst you do get full
colour pixels, and hence increased resolution over a similar pixel count
BFA camera, the response is a poor match to the eye. Foveon's highest
response is to blue, then green and then red. With this major mismatch
in spectral response between the sensor and what we perceive, the Foveon
design requires significantly higher matrix manipulation to reproduce
the visual image, with its consequential tendency to excess noise and
colour balance errors across the human visual spectrum. Sigma cameras
are notorious for their inability to get consistent flesh tones.

Kodak's proposed RGBW 2x2 matrix has similar problems, although it comes
with the benefit of improved broadband luminance response, so it isn't
all bad. Colour purity in good light isn't as good as conventional
Bayer, but low light sensitivity is much better.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

Kennedy McEwen wrote:
In article , nospam
writes
In article
,
Bubba wrote:

You didn't know that Bayer sensors have twice as many green pixels?

No. That's why I post questions here. Why in God's name would they
have twice as many?

because there are three primary colours that need to fit into a 2x2
grid.

Nothing of the sort. There is no requirement for a 2x2 grid, the
colours could be arranged as triads, as in the dots on a conventional
colour CRT, or as linear triplets as in the lines of a Trinitron screen
or LCD & Plasma displays.

They would be expensive and would reduce the number of total pixels.
It would also require interpolation to translate to the ubiquitous
retangular grid of displays and file formats.

--
Ray Fischer

In article , Alfred
Molon wrote:

Mismatched spectral response is one of the reasons why the Foveon
concept isn't as good as its supporters claim. Whilst you do get full
colour pixels, and hence increased resolution over a similar pixel count
BFA camera, the response is a poor match to the eye. Foveon's highest
response is to blue, then green and then red.

If so, wouldn't it be sufficient to multiply the blue channel by a
factor 1.0?

no

Or are you claiming that cameras with a prism and three separate sensors
(R, G and B), are a bad solution because they do not match the human
eye's sensitivity?

a 3 chip camera is rgb.

foveon isn't rgb, despite the misleading ads. the three layers have to
be converted to rgb.

Bubba wrote:
On Apr 21, 12:50 pm, Chris Malcolm wrote:
CMOS has an extra on-chip wiring overhead which matters a lot in very
tiny dense sensors, and not much on big DSLR sensors. The new
back-illuminated CMOS sensors remove that CMOS overhead (to the dark
side of the chip) and are bringing CMOS sensors into low-end P&S and
phone cameras.

Does this mean (what someone reviewing a Canon SX1 said) that the CMOS
sensors on $400--$500 P&S are "small" and (I suppose) mediocre? How
can you tell which CMOS sensor a particular camera has, and whether
it's any good?

You could read the manufacturers specification datasheet or practical
reviews of the cameras. The big thing about CMOS sensors is that they
can be *manufactured* on standard chip fab lines. CCD requires a
different more sophisticated process and until very recently gave
superior results in terms of uniformity. Now CMOS has caught up.

CMOS has had a lot more spent on it now as it allows mass produced
camera in a chip solutions for webcams and security cameras.

http://www.rps-isg.org/wordpress/?p=70

Is a reasonable basic level introduction (and refs therein).

You didn't know that Bayer sensors have twice as many green pixels?

No. That's why I post questions here. Why in God's name would they
have twice as many? I haven't bought a new digital camera in three

Because they know what they are doing! It is common knowledge!!!
You should do some proper background reading. All this stuff is well
documented. Are you too dim to use Wiki or something?

http://en.wikipedia.org/wiki/Bayer_mask

years, because of something I read here back then about my particular
problem--red channel flare--not ever improving in digital photography
unless (three years ago) you could afford a splendiferously expensive
camera ($+++Ks).

You really do need to post an *example* of this mythical "red flare" you
keep harping on about. It seems to me like you never managed to
understand how to use your original camera probably an RTFM fault given
your postings here.

Regards,
Martin Brown

nospam wrote:
In article , Ray Fischer
wrote:

Look at DPReview's discussion of Sigma's cameras. I can't provide the
link right now.

their tests show aliasing at about the same point that bayer stops
resolving.

You do know why those are two unrelated issues?

since nothing can resolve close to nyquist, you either get aliasing or
nothing, depending if there is an anti-alias filter or not.

It's not all or none.

AA filters aren't perfect, and camera makers choose different
strengths of of AA filters as different compromises between complete
absence of any aliasing ever, and permitting some aliasing. The point
of the compromise is that a weaker AA filter will capture some more
detail resolution at the cost of permitting some aliasing to appear. A
small amount of aliasing will only be noticeable in certain specific
kinds of images. So by weakening the AA filter you could get for
detail of tree leaves in shots of trees, and you wouldn't notice
aliasing. On the other hand, if you were shooting clothed portraits
with fine cloth texture that same filter might produce annoying
aliasing which was destructive of detail.

You can see such a compromise operating in TV images. Occasionally
you'll see someone wearing a certain tie or patterned weave of
clothing which throws up enough aliasing to to have weird flickering
iridescence-like effects as they move.

--
Chris Malcolm

On Thu, 22 Apr 2010 07:48:15 +0100, Martin Brown wrote:
You really do need to post an *example* of this mythical "red flare" you
keep harping on about.

I've seen an example posted. Somewhere. Maybe not in r.p.d though.
Subject was a welder in action (probably arc). Around the
highlight (and I could believe it was *very* bright, as the rest
of the image was exposed nominally) there was a square grid
of red blobs overlaying the scene, fading out away from the
highlight. There would have been on order of twenty blobs over
the height of the whole image, had they not faded.

I wondered if the effect was a double reflection, from the sensor
surface then from a lens element surface.

- Jeremy

In article , Ray Fischer
writes
Kennedy McEwen wrote:
In article , nospam
writes
In article
,
Bubba wrote:

You didn't know that Bayer sensors have twice as many green pixels?

No. That's why I post questions here. Why in God's name would they
have twice as many?

because there are three primary colours that need to fit into a 2x2
grid.

Nothing of the sort. There is no requirement for a 2x2 grid, the
colours could be arranged as triads, as in the dots on a conventional
colour CRT, or as linear triplets as in the lines of a Trinitron screen
or LCD & Plasma displays.

They would be expensive and would reduce the number of total pixels.

Not at all. The pixels can be arranged in a standard orthogonal grid
with the CFA grouping three pixels.

It would also require interpolation to translate to the ubiquitous
retangular grid of displays and file formats.

Interpolation is inherent in all CFA formats - this would be no
different.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)