From: John Sheehy on
"acl" <achilleaslazarides(a)yahoo.co.uk> wrote in
news:1175618674.275677.252260(a)n76g2000hsh.googlegroups.com:

> On Apr 3, 12:01 am, John Sheehy <J...(a)no.komm> wrote:

>> What he's saying is that the shot noise in each layer in each RGB
>> sensel will be higher by that factor, because only 1/3 of the photons
>> are captured in each level.

> But this is the same with CFAs, it's just that you absorb the extra
> photons in the filter without detecting them.

That's true as far as absolute sensitivity is concerned. The CFA filter
does not stop the wells from filling up, eventually, however, so they
basically lower the ISO rating, while maintaining full-well capacity (or
RAW saturation capacity).

>> For the Foveon used in the SD9 and 10, that is probably an
>> issue, as color has to be converted from very unsaturated RAW data.

> If you look at absorption curves with depth (given in papers put out
> by Foveon themselves), it's quite obvious why colour separation is not
> so good. Apparently also there are problems with the red channel which
> necessitates more sharpening (maybe some sort of charge diffusion
> process-but what do I know?).

I don't remember the reason for that. Most of my x3f raw samples are on
another computer that won't boot, and I haven't gotten around to pulling
the data files off it yet. I remember that the Sigma RAW data was
actually quite clean, taken literally, as good as a Canon DSLR in the
deep shadows even at ISO 800; added into a single greyscale they were
superb, nosie-wise. When converter to RGB color, though, they get
noisier and blotchier.

>> If
>> that were not an issue, however, I don't think the shot noise would
>> be much of an issue, because the Bayer shot noise, though less in
>> each sample, is bigger in height and width, making it just as strong,
>> visually. This can actually be emulated - I'll put it on my list of
>> things to try when I find lots of time.

> I don't understand this paragraph. I'll think a bit about it.

What I'm saying is that perhaps the same issues exist as with the big-
pixel/small-pixel dilemna. Co-located RGB can be considered "small-
pixel" when you approach each color plane individually, with the same
pixel-pitch. IOW, noise at the nyquist in the blue channel in co-located
RGB is at double the frequency of noise at the nyquist in the blue
channel of the CFA-filtered capture.



--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS(a)no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
From: acl on
On Apr 4, 3:23 am, John Sheehy <J...(a)no.komm> wrote:
> "acl" <achilleaslazari...(a)yahoo.co.uk> wrote
> That's true as far as absolute sensitivity is concerned. The CFA filter
> does not stop the wells from filling up, eventually, however, so they
> basically lower the ISO rating, while maintaining full-well capacity (or
> RAW saturation capacity).

Yes exactly.

>
> >> For the Foveon used in the SD9 and 10, that is probably an
> >> issue, as color has to be converted from very unsaturated RAW data.
> > If you look at absorption curves with depth (given in papers put out
> > by Foveon themselves), it's quite obvious why colour separation is not
> > so good. Apparently also there are problems with the red channel which
> > necessitates more sharpening (maybe some sort of charge diffusion
> > process-but what do I know?).
>
> I don't remember the reason for that. Most of my x3f raw samples are on
> another computer that won't boot, and I haven't gotten around to pulling
> the data files off it yet. I remember that the Sigma RAW data was
> actually quite clean, taken literally, as good as a Canon DSLR in the
> deep shadows even at ISO 800; added into a single greyscale they were
> superb, nosie-wise. When converter to RGB color, though, they get
> noisier and blotchier.

I read some papers that I found on their own site (I couldn't quickly
find them, I have them printed out but at the moment they are half a
continent away), where they showed absorption vs depth curves for
different wavelengths, and it was fairly obvious (to me) that this is
not a good way to colour separate (it's not selective enough). And if
you look at the matrix for white balance, there are large off-diagonal
components (exactly for this reason). Also if you look at the dcraw
source, you can see what kind of acrobatics is needed to get a good
image; mostly because of colour problems, I think. Especially in the
red, which is the deepest (that's why I guess it's some sort of
diffusion effect; but that's what it is, a random guess).

>
> >> If
> >> that were not an issue, however, I don't think the shot noise would
> >> be much of an issue, because the Bayer shot noise, though less in
> >> each sample, is bigger in height and width, making it just as strong,
> >> visually. This can actually be emulated - I'll put it on my list of
> >> things to try when I find lots of time.
> > I don't understand this paragraph. I'll think a bit about it.
>
> What I'm saying is that perhaps the same issues exist as with the big-
> pixel/small-pixel dilemna. Co-located RGB can be considered "small-
> pixel" when you approach each color plane individually, with the same
> pixel-pitch. IOW, noise at the nyquist in the blue channel in co-located
> RGB is at double the frequency of noise at the nyquist in the blue
> channel of the CFA-filtered capture.
>

Ah I see. Yes, it'll be higher-frequency hence better.

From: David J. Littleboy on

"John Sheehy" <JPS(a)no.komm> wrote:
>
>> I don't understand this paragraph. I'll think a bit about it.
>
> What I'm saying is that perhaps the same issues exist as with the big-
> pixel/small-pixel dilemna. Co-located RGB can be considered "small-
> pixel" when you approach each color plane individually, with the same
> pixel-pitch.

You are sounding as though you are saying the same thing that I was. The
effective size of each sensel in a 3-color system is 1/3 that of the size of
each pixel in a Bayer system.

So at the base ISO, for an exposure that puts 18% gray at 18%, the 3-color
system's red sensel has sqrt(3) worse noise than the Bayer red pixels.

> IOW, noise at the nyquist in the blue channel in co-located
> RGB is at double the frequency of noise at the nyquist in the blue
> channel of the CFA-filtered capture.

Noise and Nyquist are orthogonal, unrelated, different even.

David J. Littleboy
Tokyo, Japan


From: acl on
On Apr 4, 3:50 am, "David J. Littleboy" <davi...(a)gol.com> wrote:
> "John Sheehy" <J...(a)no.komm> wrote:
>
> >> I don't understand this paragraph. I'll think a bit about it.
>
> > What I'm saying is that perhaps the same issues exist as with the big-
> > pixel/small-pixel dilemna. Co-located RGB can be considered "small-
> > pixel" when you approach each color plane individually, with the same
> > pixel-pitch.
>
> You are sounding as though you are saying the same thing that I was. The
> effective size of each sensel in a 3-color system is 1/3 that of the size of
> each pixel in a Bayer system.
>
> So at the base ISO, for an exposure that puts 18% gray at 18%, the 3-color
> system's red sensel has sqrt(3) worse noise than the Bayer red pixels.

You are assuming that the well capacity is the same (total) but is
split in 3 for each channel. Why are you assuming this? (It may well
be true, I am not disagreeing).

>
> > IOW, noise at the nyquist in the blue channel in co-located
> > RGB is at double the frequency of noise at the nyquist in the blue
> > channel of the CFA-filtered capture.
>
> Noise and Nyquist are orthogonal, unrelated, different even.

No the dominant frequency present in the noise spectrum makes a huge
difference on a print. Higher is better (ie fine-grained noise is
better than coarse-grained).

From: David J. Littleboy on

"acl" <achilleaslazarides(a)yahoo.co.uk> wrote:
> "David J. Littleboy" <davi...(a)gol.com> wrote:
>>
>> You are sounding as though you are saying the same thing that I was. The
>> effective size of each sensel in a 3-color system is 1/3 that of the size
>> of
>> each pixel in a Bayer system.
>>
>> So at the base ISO, for an exposure that puts 18% gray at 18%, the
>> 3-color
>> system's red sensel has sqrt(3) worse noise than the Bayer red pixels.
>
> You are assuming that the well capacity is the same (total) but is
> split in 3 for each channel. Why are you assuming this? (It may well
> be true, I am not disagreeing).

Basic physics: charge storage takes space. Put three charge storage devices
in the same area, and they'll each have 1/3 the capacity. The technology
used also has a large effect, but these things are all fabricated using
essentially the same technology.

David J. Littleboy
Tokyo, Japan


First  |  Prev  |  Next  |  Last
Pages: 1 2 3 4 5 6 7 8 9
Prev: low light
Next: Bayer filter removal