The Joy of Pixel Density

Started Jul 13, 2008 | Discussions
richardplondon
richardplondon Forum Pro • Posts: 10,892
Re: Thanks John

Brad Nichol wrote:

Hi Guys

just on the stacking issue to clear things up I use it in two
different ways.

Curious: have you ever tried "enfuse"? I run it direct from Lightroom, for occasional high contrast shots that I have bracketed with a compact camera, and it does a very nice job with little fuss. This includes aligning and warping any discrepancies between the (generally handheld) shots.

Of course, if you enjoy the stacking process, and the detailed control it gives you, that's a different matter. I guess Photomerge can help if a tripod has not been used.

I like enfuse because it (can be) simple to use, while avoiding artificial-looking local contrast effects of the type often found with Photomatix and the like.

RP

 richardplondon's gear list:richardplondon's gear list
Panasonic LX10 Pentax K-5 Pentax smc DA 21mm F3.2 AL Limited Pentax smc DA 35mm F2.8 Macro Limited Pentax smc DA 70mm F2.4 AL Limited +7 more
Brad Nichol Regular Member • Posts: 117
I will look into it

Gday Richard

I have never tried enfuse though it sounds like a good idea.

Basically I stack for the total control, I always shoot tripod mounted or very firmly mounted.

Stacking can increase the detail a bit as well oddly enough, especially if there is just the tiniest bit of movement between the frames.

One thing I didn't mention is I up-rez 400% before stacking so you can imagine this is pretty taxing on processing power and RAM.

Anyhow i will look into enfuse, thanks for that.

-- hide signature --

Zero-one imaging

upnorthfar Regular Member • Posts: 492
Dynamic Range

What about DR and small pixels? Can current DSLR DR be achieved with digicam sensor density?

simpy Veteran Member • Posts: 3,090
[never mind]

text removed

 simpy's gear list:simpy's gear list
Olympus E-M1 Olympus M.Zuiko Digital 17mm F1.8 Olympus M.Zuiko Digital ED 9-18mm F4.0-5.6 Olympus M.Zuiko Digital 45mm F1.8 Olympus Zuiko Digital ED 50mm 1:2.0 Macro +7 more
hwalker Contributing Member • Posts: 961
Sensor vs. Image Processing

I've read that argument many times, but I'd like to see comparisons between images using the same raw conversion process, as opposed to the JPEG files where you're really comparing the image processing rather than the sensor. Read the DPreview's comment from the S6000 review

"(using the raw converter) Here, with both Luminance and Chroma noise reduction their defaults, you can see just how noisy the output from the (Fuji 6 megapixel sensor) is, and just how much work Fuji is doing on the files to produce the final image"

And even when you're using a raw converter, you cannot always be sure that particular raw converter you're using uses the same converting methods for all the sensors, so the testing will take a very careful consideration in choosing the converting method.

With regards to their newer modes, I suspect that Fuji is trying too hard to reduce noise at the pixel level, which actually makes the overall image quality worse. But since I don't have the cameras I'm just suspecting based on the info. available.

Right now, I am testing a few digital compacts from Sony, Nikon and Kodak, and I suspect they all use the same 1/2.5" Sony 8 mega pixel sensor. However the images they produce differ vastly in terms of noise and noise. I think the importance of image processing is highly underrated.

Amin Sabet wrote:

I can't get a high ISO
G7 or G9 print to look as good at 8x10" as my old F31 images did, and
I don't believe that Fuji has the amazing in-camera NR (otherwise
their F100fd 6MP mode would be able to match the F31 at same output
size, which it can't) nor do I believe that my G7/G9 postprocessing
skills are deficient. Furthermore, while I can see how downsizing
images can improve dynamic range, it doesn't equalize DR with that of
large pixel cameras in my practical experience.

richardplondon
richardplondon Forum Pro • Posts: 10,892
Re: [never mind]

simpy wrote:

text removed

that was my exact reaction too....

RP

 richardplondon's gear list:richardplondon's gear list
Panasonic LX10 Pentax K-5 Pentax smc DA 21mm F3.2 AL Limited Pentax smc DA 35mm F2.8 Macro Limited Pentax smc DA 70mm F2.4 AL Limited +7 more
Bogdan_ Regular Member • Posts: 116
Re: Subscribing to thread...[nt]

Me to.

Tristan Cope Senior Member • Posts: 1,141
Re: Density limits

I think shot noise is a significant problem when the photon count per pixel is very low. And binning won't help.

-- hide signature --

I just like the pictures.

 Tristan Cope's gear list:Tristan Cope's gear list
Sony RX100 Canon EOS 5D Mark III Sigma SD14 Panasonic Lumix DMC-GX1 Panasonic Lumix DMC-GX7 +1 more
Tristan Cope Senior Member • Posts: 1,141
Shot noise

This is what I was trying to get at with my post. If the photon count for each individual pixel gets very low, then then signal to noise ration dramatically falls.

Let say, for example that we have a 40MP 1/1.8 sensor, and that in certain light conditions the average photon count per pixel is aound 1. Because of the effect of shot noise the signal will be indistinguishable from the noise. Binning pixels won't help. Sufficient photons per pixel are required to statistically distinguish signal from noise.

This would still be the case even if there was no read noise.

-- hide signature --

I just like the pictures.

 Tristan Cope's gear list:Tristan Cope's gear list
Sony RX100 Canon EOS 5D Mark III Sigma SD14 Panasonic Lumix DMC-GX1 Panasonic Lumix DMC-GX7 +1 more
aclo Regular Member • Posts: 480
Re: Density limits

Tristan Cope wrote:

I think shot noise is a significant problem when the photon count per
pixel is very low. And binning won't help.

Shot noise is always present, but is indeed more significant at low intensities (it goes like the square root of the signal). And binning most definitely helps... bin nxn pixels, the signal to noise ratio due to shot noise will go up by a factor n. [because the signal grows n^2 times, while the squares of the n^2 noise magnitudes should be added and then the square root taken, since we're dealing with uncorrelated sources, hence central limit theorem blah blah].

It's not hard to see why even without knowing how to combine uncorrelated noise sources: this shot noise is really part of the signal itself.

Where did you get the idea that binning doesn't help?

aclo Regular Member • Posts: 480
Re: Shot noise

Tristan Cope wrote:

This is what I was trying to get at with my post.

JPS' post is actually in agreement with the one I made... Can you explain what you had in mind a bit better?

If the photon count
for each individual pixel gets very low, then then signal to noise
ration dramatically falls.

Let say, for example that we have a 40MP 1/1.8 sensor, and that in
certain light conditions the average photon count per pixel is aound
1. Because of the effect of shot noise the signal will be
indistinguishable from the noise. Binning pixels won't help.

The last statement isn't true. Can you explain the thinking that leads to it?

Sufficient photons per pixel are required to statistically
distinguish signal from noise.

This would still be the case even if there was no read noise.

Certainly not (although I do agree it seems like what you say should be true, at first sight).

GordonBGood Veteran Member • Posts: 6,308
Re: Shot noise

Tristan Cope wrote:

This is what I was trying to get at with my post. If the photon count
for each individual pixel gets very low, then then signal to noise
ration dramatically falls.

Let say, for example that we have a 40MP 1/1.8 sensor, and that in
certain light conditions the average photon count per pixel is aound
1. Because of the effect of shot noise the signal will be
indistinguishable from the noise. Binning pixels won't help.
Sufficient photons per pixel are required to statistically
distinguish signal from noise.

This would still be the case even if there was no read noise.

No, binning photosites would help. For the situation you describe above, by binning four photosites together the total number of electrons would go to four but the noise by the square root of the number of electrons would be two, so you would have a 2:1 Signal to Noise Ratio (SNR) or twice as good as not binning. I am not considering read noise, just as you say.

Regards, GordonBGood

Tristan Cope Senior Member • Posts: 1,141
Re: Shot noise

aclo wrote:

Tristan Cope wrote:

If the photon count
for each individual pixel gets very low, then then signal to noise
ration dramatically falls.

Let say, for example that we have a 40MP 1/1.8 sensor, and that in
certain light conditions the average photon count per pixel is aound
1. Because of the effect of shot noise the signal will be
indistinguishable from the noise. Binning pixels won't help.

The last statement isn't true. Can you explain the thinking that
leads to it?

Yes. Lets take the extreme case and say that one photon is falling on each sensor element. Signal = 1, noise = sqrt 1 = 1. Signal and noise are equal (essentially photon flux per sensor element is entirely random regardless of the signal). This is a quantum effect.

"Binning can be done in firmware after signal readout and amplification, but this wouldn't reduce shot noise. To bin sensor elements together in a way that reduced shot noise would require this to be done in hardware at sensor level before readout. 2x2 (or 4x4) blocks blocks of immediately adjacent sensor elements would have to be combined before readout so they were treated as a single large sensor element (i.e. a bigger "pixel"). This can't be done with CFA sensors due to the CFA array. You would be binning pixels with different colour information.

I don't think on-sensor hardware binning is something that can be easily accomplished with CFA sensors (unlike monochrome sensors). If it was, all the manufacturers would be doing it. Here is Kodaks latest effort using panchromatic elements:
http://johncompton.pluggedin.kodak.com/default.asp?item=624876

Is there any CFA camera on the market currently that claims to use hardware pixel binning?

-- hide signature --

I just like the pictures.

 Tristan Cope's gear list:Tristan Cope's gear list
Sony RX100 Canon EOS 5D Mark III Sigma SD14 Panasonic Lumix DMC-GX1 Panasonic Lumix DMC-GX7 +1 more
Tristan Cope Senior Member • Posts: 1,141
See below [NT]

see below
--
I just like the pictures.

 Tristan Cope's gear list:Tristan Cope's gear list
Sony RX100 Canon EOS 5D Mark III Sigma SD14 Panasonic Lumix DMC-GX1 Panasonic Lumix DMC-GX7 +1 more
simpy Veteran Member • Posts: 3,090
binning always helps

Tristan Cope wrote:

Yes. Lets take the extreme case and say that one photon is falling on
each sensor element. Signal = 1, noise = sqrt 1 = 1. Signal and noise
are equal (essentially photon flux per sensor element is entirely
random regardless of the signal). This is a quantum effect.

Even if the average signal is much less than a single photon, and you couldn't say anything about the signal based on a single pixel, binning would still help. It works because we know that the shot noise is randomly distributed with a mean of zero, and the signal isn't. Binning over enough pixels can be used to reduce the noise by an arbitrary factor, so you can always get a high signal/noise ratio, albeit at lower resolution.

Simon

 simpy's gear list:simpy's gear list
Olympus E-M1 Olympus M.Zuiko Digital 17mm F1.8 Olympus M.Zuiko Digital ED 9-18mm F4.0-5.6 Olympus M.Zuiko Digital 45mm F1.8 Olympus Zuiko Digital ED 50mm 1:2.0 Macro +7 more
richardplondon
richardplondon Forum Pro • Posts: 10,892
OT: enfuse

http://forums.dpreview.com/forums/read.asp?forum=1036&message=27585099&q=richardplondon+enfuse&qf=m

I put up an example here (though not the most interesting shot)

RP

 richardplondon's gear list:richardplondon's gear list
Panasonic LX10 Pentax K-5 Pentax smc DA 21mm F3.2 AL Limited Pentax smc DA 35mm F2.8 Macro Limited Pentax smc DA 70mm F2.4 AL Limited +7 more
GordonBGood Veteran Member • Posts: 6,308
Re: Shot noise

As follows:

Tristan Cope wrote:

aclo wrote:

Tristan Cope wrote:

If the photon count
for each individual pixel gets very low, then then signal to noise
ration dramatically falls.

Let say, for example that we have a 40MP 1/1.8 sensor, and that in
certain light conditions the average photon count per pixel is aound
1. Because of the effect of shot noise the signal will be
indistinguishable from the noise. Binning pixels won't help.

The last statement isn't true. Can you explain the thinking that
leads to it?

Yes. Lets take the extreme case and say that one photon is falling on
each sensor element. Signal = 1, noise = sqrt 1 = 1. Signal and noise
are equal (essentially photon flux per sensor element is entirely
random regardless of the signal). This is a quantum effect.

"Binning can be done in firmware after signal readout and
amplification, but this wouldn't reduce shot noise. To bin sensor
elements together in a way that reduced shot noise would require this
to be done in hardware at sensor level before readout. 2x2 (or 4x4)
blocks blocks of immediately adjacent sensor elements would have to
be combined before readout so they were treated as a single large
sensor element (i.e. a bigger "pixel"). This can't be done with CFA
sensors due to the CFA array. You would be binning pixels with
different colour information.

I don't think on-sensor hardware binning is something that can be
easily accomplished with CFA sensors (unlike monochrome sensors). If
it was, all the manufacturers would be doing it. Here is Kodaks
latest effort using panchromatic elements:
http://johncompton.pluggedin.kodak.com/default.asp?item=624876

Is there any CFA camera on the market currently that claims to use
hardware pixel binning?

Hardware binning is not necessary as software binning does the same thing. The shot noise will be purely randomly distribute and thus when one takes an average over a large number of 2 X 2 software binned photosites, the shot noise will be reduced to the square root of 4 = 2 just as if it was hardware binned.

This software binning can be done with CFA sensors by combining like coloured photosites just as easily as it can be done for a black and white sensor in that the individual photosites are not co-dependent (or shouldn't be if designed properly anyway).

Yes, individual single electron capacities would be a quantum effect, but an average over a reasonably number of such quantum photosites would look more and more like the classical case.

Regards, GordonBGood

aclo Regular Member • Posts: 480
Re: Shot noise

Tristan Cope wrote:

aclo wrote:

Tristan Cope wrote:

If the photon count
for each individual pixel gets very low, then then signal to noise
ration dramatically falls.

Let say, for example that we have a 40MP 1/1.8 sensor, and that in
certain light conditions the average photon count per pixel is aound
1. Because of the effect of shot noise the signal will be
indistinguishable from the noise. Binning pixels won't help.

The last statement isn't true. Can you explain the thinking that
leads to it?

Yes. Lets take the extreme case and say that one photon is falling on
each sensor element. Signal = 1, noise = sqrt 1 = 1. Signal and noise
are equal (essentially photon flux per sensor element is entirely
random regardless of the signal). This is a quantum effect.

That the signal and noise have the same magnitude is irrelevant, as I explained before (and others do later). And it also happens with classical objects being randomly deposited onto a surface (such as epitaxial growth), so it's not quantum (except inasmuch as the photons being discrete is a "quantum" effect). Anyway this is off topic.

OK think of this: imagine a photosite split in 10x10 smaller squares. Suppose I "detect" photons separately in each, and there is no read noise. Do you not see that shot noise will be unaffected by whether I a) write out all 100 values, then add them, or b) I have a computer (or analog counter) do it? (think about this a bit before dismissing it). [and the photons aren't coherent so we really can argue like above].

I hope the idea gets across with my example.

"Binning can be done in firmware after signal readout and
amplification, but this wouldn't reduce shot noise. To bin sensor
elements together in a way that reduced shot noise would require this
to be done in hardware at sensor level before readout. 2x2 (or 4x4)
blocks blocks of immediately adjacent sensor elements would have to
be combined before readout so they were treated as a single large
sensor element (i.e. a bigger "pixel"). This can't be done with CFA
sensors due to the CFA array. You would be binning pixels with
different colour information.

Yes, I agree that the presence of the CFA complicates things and makes all the discussions here not directly applicable. You can bin after demosaicing, but the demosaicing process introduces long-range correlations in the noise, and this decreases the effectiveness of binning.

That's a different story from the noise and signal being equal, though.

Is there any CFA camera on the market currently that claims to use
hardware pixel binning?

I don't think so, and it wouldn't make much sense, as you say. Even worse, just binning 2x2 and so on would cause aliasing, so things are even worse in practice.

dmanthree
dmanthree Veteran Member • Posts: 7,759
Re: The Joy of Pixel Density

So you're trying to prove that given the same sensor area, a higher number of pixels will produce a better image? Using two vastly different cameras and up-sampling one to match the other? Huh?

This is like trying to determine the fuel efficiency of motors with hundreds of cylinders against the same size engine with a few cylinders, except you've compared a Smart motor with a locomotive deisel.

I know this has been posted and discussed in usenet, but the result is the same: this is a test with no pratical value since you can't really perform it correctly, and the results do not reflect any sort of diminishing returns in either direction. For example, what kind of image would be produced by a 100M sensor that's only 1/2.5"? Anything usable? At what sensitivity?
--
-------------------------------------------------

OP John Sheehy Forum Pro • Posts: 21,743
Re: Density limits

Tristan Cope wrote:

I think shot noise is a significant problem when the photon count per
pixel is very low. And binning won't help.

You'd better think about that just a little bit longer. This is the easiest myth to disprove.

Imagine you have 1 box, 2 inches square. You have 4 boxes, 1 inch square, together in a 2" x 2" bigger square, like the bigger box. In the little boxes you have 9, 3, 2, and 0 marbles. In the large box you have 14 marbles.

Which has more noise? Which has more resolution?

-- hide signature --

John

Keyboard shortcuts:
FForum MMy threads