From: acl on
On Mar 16, 6:50 pm, John Sheehy <J...(a)no.komm> wrote:
> "acl" <achilleaslazari...(a)> wrote innews:1173975897.845961.131720(a)
> > On Mar 15, 2:31 pm, John Sheehy <J...(a)no.komm> wrote:
> >> that standard deviation is only *one* factor in the noise equation;
> >> magnification is another, and the low noise of big pixels is visually
> >> magnified when the pixels are magnified along with the subject.
> > Exactly, and if you don't need the extra pixels you can bin.
> Yes, but I think it is very important to stress that there is no need to
> bin to get the benefit. Often, the situation is described in such a way
> that binning or downsampling are *necessary* to get the benefit. There
> is a cult of pixel-for-pixel's sake that misses the forest (image or
> subject) for the trees (pixels), IMO. To me, the most important quality
> factors in descending importance are:
> 1) SQ (subject quality)
> 2) IQ (image quality)
> 3) PQ (pixel quality)
> If #3 doesn't also help #1, it is in vein.
> >> 2) slight benefit in photons collection rate per unit of sensor area
> >> due
> >> to less wasted space on the sensor (not always realized, however; my
> >> 1.97u FZ50, for example, collects about the same number of photons per
> >> unit of area as the 1DmkII, at RAW saturation for the same ISO).
> > Well, as long as there are no constant noise sources (eg 10 electrons/
> > pixel independent of the area). I have no idea if there are or not.
> Blackframe read noise on my FZ50 is about 3.34 electrons at ISO 100 (4800
> electrons at saturation), and about 2.71 electrons at ISO 1600 (about 300
> electrons at saturation). Binned down 3x3 (to DSLR size), that's about
> 0.9 and 1.11 electrons (43,200 and 2700 max), respectively.

I don't understand, why are you dividing by three? for noise r, you
want sqrt(9*r^2)=3r, ie around 10 and 9 electrons or so. Or am I
missing something?

> These values are derived from multiplying the standard deviation of FZ50
> blackframes by 1.66, since the Panasonic RAW files, unfortunately, are
> clipped at the blackpoint, a very bad idea that most camera manufacturers
> engage in. Canon is one of the few that leave a bias in the RAW data
> with a full symmetrical noise histogram with positive and negative noise.

Yes I think Nikon does this too I didn't check in detail though, just
an impression I got from looking at blackframes in IRIS some time ago;
but maybe it is also dcraw's handling of the data, I don't know. The
standard deviation of the noise was slightly higher at slightly larger
raw values (5 or so) than at 0; so I thought that it is simply
clipping the left edge of the noise distribution. But I didn't
specifically check.

> Clipping at black means bright, noise-clouded blacks with minute signals
> clipped. Blackpoint clipping should not occur, IMO, until your RAW
> conversion needs to enter a gamma-adjusted state (where negative signal
> or noise are meaningless). All white-balancing, resampling, and
> demosaicing should be done before the clipping, to maintain black blacks,
> and a minimum of noise. I would guess that most converters don't
> maintain this state, but clip Canons at black immediately upon loading
> RAW data to get it in the same state as most other cameras. In my own
> hand-conversions in IRIS and in PS with filtermeister and RAW linear
> greyscale sources, I have made conversions with much less color tint and
> bright haze in the deepest shadows. In fact, I have even promoted images
> to a higher bit-depth before any interpolative actions, for more
> precision. RAW converters, generally, are taking a lot of short-cuts,
> IMO, and are not delivering what they can in these areas. They seem
> focused mainly on the post-processing values like skin color and tonal
> curves in the highlight areas, etc.

I must admit that I cannot imagine that this clipping will have any
serious consequences, except maybe if you do binning (it'll lighten
the black), but I'm not sure how important it would be. But maybe if
it's really noise then it will have an effect. I can't check, though.
Could you post examples with and without this when you get time? I am

From: acl on
On Mar 16, 9:18 pm, John Sheehy <J...(a)no.komm> wrote:
> "acl" <achilleaslazari...(a)> wrote innews:1174058129.036168.91300(a)
> > But personally I hope this kind of noise could go down enough so we
> > can have high pixel density sensors which will give more flexibility
> > in trading off noise for resolution. Hopefully with built-in binning
> > for the raw files too, as always having 60MB raw files seems a bit
> > wasteful (but the again 300KB for an executable seemed huge to me in
> > 1988).
> The smaller the pixels become, the less bit depth you need to record them
> at the same level of accuracy.

Yes you're right. If the full well capacity is proportional to pixel
area (and it must be something like that), then the size of a raw file
should in principle remain roughly constant as we decrease the pixel
size (until edge effects take over and it is not useful to miniaturise
any more due to the fill factor not being 1).

Nice to know that we won't be needing 100GB cards soon.

From: acl on
On Mar 16, 9:03 pm, John Sheehy <J...(a)no.komm> wrote:
> Sorry, I had just woken up after an 18-hour emergency shift at work.
> I divided by three when I should have multiplied by three. I was thinking
> in terms of the noise-to-signal ratio, and applied it to absolute photon
> counts.
> That should have read:
> Blackframe read noise on my FZ50 is about 3.34 electrons at ISO 100
> (4800 electrons at saturation), and about 2.71 electrons at ISO 1600
> (about 300 electrons at saturation). Binned down 3x3 (to DSLR size),
> that's about 8.1 and 10 electrons (43,200 and 2700 max),
> respectively.

OK, I missed this post before, that explains it. Thanks.

From: John Sheehy on
Lionel <usenet(a)> wrote in

> On Fri, 16 Mar 2007 12:29:52 +0900, "David J. Littleboy"
> <davidjl(a)> wrote:
>>I think you guys are talking past each other here.
>>I think John is arguing that _for a sensor of a given size_, larger
>>pixels aren't any better.
> But they /are/ better! - That's why the sensor designers are
> constantly trying to improve the fill-factor, ie; make the pixels (or,
> more accurately, the actual photo diode surface, which is smaller than
> the pixel size) bigger for a given sensor size/resolution ratio. This
> is because the bigger the suface of the photodiode (as a proportion of
> the size of that pixel on the sensor), the more photons it'll collect
> for a given exposure. And, all else being equal, more photons equals a
> better signal to noise ratio.

Bigger pixels capture more photons per *PIXEL*; not per unit of sensor
area, which is much more relevant. Noise per pixel is just a myopic,
"missing-the-forest-for-the-trees" academic curiosity with no direct
relationship to practical photography, in terms of pixel density.

Think hard about this, if you think large pixels give better images, for
shot noise reasons:

Imagine that you had 16 square containers, and you gave them to an
assistant, to place in a tight 4x4 array out in a field, to measure
rainfall during a certain period of time.

Your assistant is me, and I decide to replace the 16 containers with 64
square containers, 4 of which fit in the same space as 1 of yours. I come
back to you with a list of results from 64 smaller containers, instead of
the 16 you asked for. The list is longer, and the total count is the
same as it would be if I used your original containers, but have I
created any *NOISE*?

Of course not, but many people believe so for capturing photons instead
of raindrops!

The fact is, the smaller pixels give you more detail about where the
raindrops (or photons) fell. That is not *NOISE*.


<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS(a)no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
From: Pat on
On Mar 16, 2:38 pm, John Sheehy <J...(a)no.komm> wrote:
> "Pat" <gro...(a)> wrote innews:1174068774.683691.273470(a)
> > So if you don't like my idea -- which I'm not sure I even like my idea
> > -- what dSLR would you suggest for the OP with a $1000 budget (don't
> > forget sales tax and shipping)?
> A used 20D or a used Rebel XT.
> With lenses, might run a little bit more than a used 300D, but will be far
> faster in operation, and better in low light. Sometimes it's worth going a
> little over budget to get something a notch or two better.
> --
> <>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
> John P Sheehy <J...(a)no.komm>
> ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><

I don't necessarily disagree with that, but then $1000 would really be
the budget then, would it.