The death of the Bayer filter? Maybe not.

I don't hate the Bayer. I think it's a clever way to deal with a problem
that some find objectionable. But its days could be numbered.
Don't know if resolution is improved yet, but it's nice to know that it
only costs $1700(!) to get a non-flawed Fuji when mainstream makers produce
$500 DSLRs that meet the spec.

http://www.dpreview.com/articles/612...ons-using-the-
fujifilm-x-pro1/6

On 4/04/2012 5:30 p.m., Mxsmanic wrote:
Rich writes:

I don't hate the Bayer. I think it's a clever way to deal with a problem
that some find objectionable. But its days could be numbered.
Don't know if resolution is improved yet, but it's nice to know that it
only costs $1700(!) to get a non-flawed Fuji when mainstream makers produce
$500 DSLRs that meet the spec.

A Bayer filter and an anti-aliasing filter are two different things.
IIRC this camera uses a "pseudo random" pattern for RGBG filters, which
theoretically gets rid of *colour moire, and possibly reduces luminance
moire (by more random placement of green sensels than a repeating bayer
pattern), so there's less need for a low pass filter.
I guess that demosaicking raw files is rather complicated, the only raw
converter supporting the camera for now is Silkypix - which I've tried
in the past and I thought it really sucked, but YMMV.
It's a good idea, but with higher and higher resolution with
conventional bayer filters, when at those higher resolutions the low
pass filter needs to be less aggressive (on a whole image basis) to
produce the same anti-aliasing effect (on a per pixel basis as a lower
resolution sensor), any "need" for such an innovative solution may have
already passed.

*It can't work perfectly, only reduce the incidence perhaps in repeating
patterns - as any point of white light hitting the sensor at a size less
than one sensel, is going to be recorded as R, G, or B - not white.

Me wrote:
On 4/04/2012 5:30 p.m., Mxsmanic wrote:
Rich writes:

I don't hate the Bayer. I think it's a clever way to deal with a problem
that some find objectionable. But its days could be numbered.
Don't know if resolution is improved yet, but it's nice to know that it
only costs $1700(!) to get a non-flawed Fuji when mainstream makers produce
$500 DSLRs that meet the spec.

A Bayer filter and an anti-aliasing filter are two different things.
IIRC this camera uses a "pseudo random" pattern for RGBG filters, which
theoretically gets rid of *colour moire, and possibly reduces luminance
moire (by more random placement of green sensels than a repeating bayer
pattern), so there's less need for a low pass filter.

It's nothing like pseudo-random.
http://www.dpreview.com/previews/fujifilmxpro1/3 shows that it's a 6x6
array instead of a 2x2 array. This 6x6 array is rotationally
symmetrical, and it repeats across the sensor. It has the interesting
property that there are fewer blue and red pixels than a Bayer sensor
has: 8R:20G:8B instead of 9:18:9. It probably won't make much
difference.

Andrew.

Mxsmanic writes:

Me writes:

IIRC this camera uses a "pseudo random" pattern for RGBG filters, which
theoretically gets rid of *colour moire, and possibly reduces luminance
moire (by more random placement of green sensels than a repeating bayer
pattern), so there's less need for a low pass filter.

It would also reduce color resolution. If it's actually being done, it sounds
like a gimmick.

No, it wouldn't. It has the same proportions of pixels in each color,
the placement is simply less regular -- more like film grain, less like
a tic-tac-toe board.

The whole issue of aliasing is past history; I don't know why people still
worry about it. When you have a large number of pixels, you don't need to
worry about aliasing. So the ultimate solution for aliasing is more pixels.
There will always be aliasing of details that are sufficiently small, if the
lens can resolve them, but the aliasing will not be noticeable or
troublesome.

I've seen recent examples of pictures containing horrible aliasing.

Even in the old days, when sensors had far fewer pixels, I never really
noticed problems with aliasing. I'd rather take the risk of aliasing than put
up with the blur of anti-aliasing.

You mostly haven't had that choice. And if you're working fast and in
field conditions, you're MUCH better off with the AA filter.
--
David Dyer-Bennet, ; http://dd-b.net/
Snapshots: http://dd-b.net/dd-b/SnapshotAlbum/data/
Photos: http://dd-b.net/photography/gallery/
Dragaera: http://dragaera.info

On 5/04/2012 1:31 a.m., Andrew Haley wrote:
wrote:
On 4/04/2012 5:30 p.m., Mxsmanic wrote:
Rich writes:

I don't hate the Bayer. I think it's a clever way to deal with a problem
that some find objectionable. But its days could be numbered.
Don't know if resolution is improved yet, but it's nice to know that it
only costs $1700(!) to get a non-flawed Fuji when mainstream makers produce
$500 DSLRs that meet the spec.

A Bayer filter and an anti-aliasing filter are two different things.
IIRC this camera uses a "pseudo random" pattern for RGBG filters, which
theoretically gets rid of *colour moire, and possibly reduces luminance
moire (by more random placement of green sensels than a repeating bayer
pattern), so there's less need for a low pass filter.

It's nothing like pseudo-random.
http://www.dpreview.com/previews/fujifilmxpro1/3 shows that it's a 6x6
array instead of a 2x2 array. This 6x6 array is rotationally
symmetrical, and it repeats across the sensor. It has the interesting
property that there are fewer blue and red pixels than a Bayer sensor
has: 8R:20G:8B instead of 9:18:9. It probably won't make much
difference.

I hadn't seen that on DPReview. I thought I'd seen a description on
Fuji's website describing the array pattern as semi-random. It looks
like it's a lot more pseudo and a lot less random than I'd expected.

On 5/04/2012 2:58 a.m., RichA wrote:
On Apr 4, 9:31 am, Andrew
wrote:
wrote:
On 4/04/2012 5:30 p.m., Mxsmanic wrote:
Rich writes:

I don't hate the Bayer. I think it's a clever way to deal with a problem
that some find objectionable. But its days could be numbered.
Don't know if resolution is improved yet, but it's nice to know that it
only costs $1700(!) to get a non-flawed Fuji when mainstream makers produce
$500 DSLRs that meet the spec.

A Bayer filter and an anti-aliasing filter are two different things.
IIRC this camera uses a "pseudo random" pattern for RGBG filters, which
theoretically gets rid of *colour moire, and possibly reduces luminance
moire (by more random placement of green sensels than a repeating bayer
pattern), so there's less need for a low pass filter.

It's nothing like pseudo-random.http://www.dpreview.com/previews/fujifilmxpro1/3shows that it's a 6x6
array instead of a 2x2 array. This 6x6 array is rotationally
symmetrical, and it repeats across the sensor. It has the interesting
property that there are fewer blue and red pixels than a Bayer sensor
has: 8R:20G:8B instead of 9:18:9. It probably won't make much
difference.

Andrew.

Didn't Sony produce a sensor with yellow pixels at one point in a
P&S? Anyone know how that turned out?
IIRC that was white pixels - but then again nothing would surprise me.
Sharp make TV panels with yellow pixels. This seems to be 99% BS.
Sometimes competition/marketing ends up driving complex and even elegant
solutions to problems which never existed.

Me writes:

On 5/04/2012 1:31 a.m., Andrew Haley wrote:
wrote:
On 4/04/2012 5:30 p.m., Mxsmanic wrote:
Rich writes:

I don't hate the Bayer. I think it's a clever way to deal with a problem
that some find objectionable. But its days could be numbered.
Don't know if resolution is improved yet, but it's nice to know that it
only costs $1700(!) to get a non-flawed Fuji when mainstream makers produce
$500 DSLRs that meet the spec.

A Bayer filter and an anti-aliasing filter are two different things.
IIRC this camera uses a "pseudo random" pattern for RGBG filters, which
theoretically gets rid of *colour moire, and possibly reduces luminance
moire (by more random placement of green sensels than a repeating bayer
pattern), so there's less need for a low pass filter.

It's nothing like pseudo-random.
http://www.dpreview.com/previews/fujifilmxpro1/3 shows that it's a 6x6
array instead of a 2x2 array. This 6x6 array is rotationally
symmetrical, and it repeats across the sensor. It has the interesting
property that there are fewer blue and red pixels than a Bayer sensor
has: 8R:20G:8B instead of 9:18:9. It probably won't make much
difference.

I hadn't seen that on DPReview. I thought I'd seen a description on
Fuji's website describing the array pattern as semi-random. It looks
like it's a lot more pseudo and a lot less random than I'd expected.

That was my reaction when reading the press release and then seeing the
diagram of the actual filter arrangement, yes.
--
David Dyer-Bennet, ; http://dd-b.net/
Snapshots: http://dd-b.net/dd-b/SnapshotAlbum/data/
Photos: http://dd-b.net/photography/gallery/
Dragaera: http://dragaera.info

Mxsmanic wrote:
Me writes:

IIRC this camera uses a "pseudo random" pattern for RGBG filters, which
theoretically gets rid of *colour moire, and possibly reduces luminance
moire (by more random placement of green sensels than a repeating bayer
pattern), so there's less need for a low pass filter.

It would also reduce color resolution. If it's actually being done, it sounds
like a gimmick.

The whole issue of aliasing is past history; I don't know why people still
worry about it. When you have a large number of pixels, you don't need to
worry about aliasing. So the ultimate solution for aliasing is more pixels.
There will always be aliasing of details that are sufficiently small, if the
lens can resolve them, but the aliasing will not be noticeable or troublesome.

Even in the old days, when sensors had far fewer pixels, I never really
noticed problems with aliasing. I'd rather take the risk of aliasing than put
up with the blur of anti-aliasing.

I started my digital photography processing with 128 x 129 pixel
images in a lab. Aliasing was a very serious problem! But accuracy was
sufficiently important that aliasing was dealt with by trying to
recognise the artefacts and the probable underlying reality (such as a
staircased stright edge) rather than trying to avoid aliasing by some
kind of blurring.

--
Chris Malcolm

Mxsmanic writes:

David Dyer-Bennet writes:

No, it wouldn't. It has the same proportions of pixels in each color,
the placement is simply less regular -- more like film grain, less like
a tic-tac-toe board.

Less regular means more anomalies in the recording of color. Yes, that's like
grain ... but is grain what you want?

Not necessarily visible anomalies, though.

I've seen recent examples of pictures containing horrible aliasing.

If there is horrible aliasing, it can be adjusted in post. That's way
preferable to blurring every single photo just on the off chance that there
might be a problem with aliasing.

Nope, horrible aliasing (generally moire) is really terribly hard to
adjust in post. The good solution is to work slowly and carefully so
you catch it immediately; some minor adjustment to the shot will
generally eliminate it (this from multiple people working regularly with
medium-format backs lacking AA filters).

You mostly haven't had that choice. And if you're working fast and in
field conditions, you're MUCH better off with the AA filter.

Nope, I want the output of the camera to be as unmanipulated as possible, so
that I can make any necessary adjustments in post. Changes made in the camera
cannot be undone, and if they ruin a photo, you're out of luck.

You reall want an AA filter for that. On balance it's the right
tradeoff for essentially any fast-moving photography excecpt perhaps
certain kinds of wildlife.

That's why so many DSLRs actually produce poor video, incidentally.

Strange how well the professionals like the output.
--
David Dyer-Bennet, ; http://dd-b.net/
Snapshots: http://dd-b.net/dd-b/SnapshotAlbum/data/
Photos: http://dd-b.net/photography/gallery/
Dragaera: http://dragaera.info

Mxsmanic writes:

Andrew Haley writes:

It's nothing like pseudo-random.
http://www.dpreview.com/previews/fujifilmxpro1/3 shows that it's a 6x6
array instead of a 2x2 array. This 6x6 array is rotationally
symmetrical, and it repeats across the sensor. It has the interesting
property that there are fewer blue and red pixels than a Bayer sensor
has: 8R:20G:8B instead of 9:18:9. It probably won't make much
difference.

I prefer more blue and red, not less. I'd like an equal number of green, red,
and blue pixels, in sufficiently high number to make favoritism for green a
moot point.

You're probably not going to get it; because it doesn't match the
behavior of the human eye very well, so it'll look less good than a less
"fair" design does (to human eyes).
--
David Dyer-Bennet, ; http://dd-b.net/
Snapshots: http://dd-b.net/dd-b/SnapshotAlbum/data/
Photos: http://dd-b.net/photography/gallery/
Dragaera: http://dragaera.info