• Please use real names.

    Greetings to all who have registered to OPF and those guests taking a look around. Please use real names. Registrations with fictitious names will not be processed. REAL NAMES ONLY will be processed

    Firstname Lastname

    Register

    We are a courteous and supportive community. No need to hide behind an alia. If you have a genuine need for privacy/secrecy then let me know!
  • Welcome to the new site. Here's a thread about the update where you can post your feedback, ask questions or spot those nasty bugs!

Trade-offs between high speed and image quality versus high resolution

John Sheehy

New member
We may be at a junction between high speed and image quality versus high resolution at the expense of some inherent DR quality. We'll see what the future has in store.

I think the future is going to say that all this pixel-centric DR/Noise dialogue is just noise.

A Panasonic FZ50 with it's pixels binned down to 1DmkII size has slightly less shot noise at all levels, and 60% of the read noise, for up to 0.7 stops more DR at the output pixel level (I don't believe you need to bin to get the benefit, though).

I really don't see how people can believe that small pixels are bad. It's a long way down past 8 microns where you will start losing photons, even with current technology. Canon's 1.3 and 1.0 sensors give good images because of their size; not because of the size of the pixels.

One shouldn't look at everything at "100% pixel view", unless they *really* know what that means. Noise has width and height, in addition to depth.
 

John Sheehy

New member
I cannot speak from my own experience since I never ever used FZ50, but here's a couple of samples from Phil's DPReview site.

Since the images are not mine I'm posting the links only.

So, you were saying...? ;-)

I said what I said, and it was correct.

What you are linking to has absolutely nothing to do with what I said.

Read what I wrote again, s l o w l y. All of the words are there, for a reason.
 
I said what I said, and it was correct.

What you are linking to has absolutely nothing to do with what I said.

Read what I wrote again, s l o w l y. All of the words are there, for a reason.


First, you're getting rude and I do not like it. I politely suggest you change the tone.

Second, the sample image from FZ50 shows way more noise than the image taken by 1DII at a similar ISO level. You may be correct in theory but my customers would not be interested in one. They need a final image. And unless we're talking some artsy exceptions, noise=bad.

Cheers!
 

Asher Kelman

OPF Owner/Editor-in-Chief
I think the future is going to say that all this pixel-centric DR/Noise dialogue is just noise.

A Panasonic FZ50 with it's pixels binned down to 1DmkII size has slightly less shot noise at all levels, and 60% of the read noise, for up to 0.7 stops more DR at the output pixel level (I don't believe you need to bin to get the benefit, though).

I really don't see how people can believe that small pixels are bad. It's a long way down past 8 microns where you will start losing photons, even with current technology. Canon's 1.3 and 1.0 sensors give good images because of their size; not because of the size of the pixels.

One shouldn't look at everything at "100% pixel view", unless they *really* know what that means. Noise has width and height, in addition to depth.

John,

You've made that statement previously and it's something I always like to hear, a sort of David v. Goliath contest. However, could you give some practical examples where the FZ50 could replace the 1DII?

Are you suggesting that the Canon's ISO range and noise is matched by the FZ50. From what I have seen the fZ50 has smeared small prints starting from ISO 400. That at least according to Phil Askey's review.

So what is the arena where you say the FZ50 trumps the 1DII as far as noise and DR.

Asher
 

Asher Kelman

OPF Owner/Editor-in-Chief
Thanks, John, I got that from the outset. O.K., now what do we do with this? IOW, here do we go with this is a practical sense? What pictures in what industry or purpose would that Panasonic Camera then trounce the 1DII. This question is asked not as a confrontation. I believe that there must be a zone where the Z50 excels. But where is it in practical terms?

Asher
 

Aaron Strasburg

New member
John,

Noise and DR per unit of sensor area? That's an unusual unit.

Perhaps what you're trying to say is that the 1DsII (or even the rumored 22MP 1DsIII) are not the end of the line for pixel density, as demonstrated by the FZ50? That's an interesting argument, if it's what you are in fact saying.

I don't think there are too many situations where the FZ50 will capture a better image than a 1DsII, but I suspect you're correct that there's yet mileage in the CMOS sensor beyond the 1DsIII.

Aaron
 

Asher Kelman

OPF Owner/Editor-in-Chief
I too had the sense that we were talking potential not practicality. Still, I wonder where the FZ50 is king?

In the meanwhile, I waver between going after the 1DIII or just improving my large format capability. I have taken out my cameras and am looking at new lenses.

For my needs, the 1DII is fine for people shots. Its landscape and building that I have issues with. There I want lots of detail, so the 1DIII does not offer to solve what I am looking for. Still I find the camera so enticing with the ability to keep itself cleaner and reach out in the lowest light, that I'm sorely tempted.

For my people work, the pixel count doesn't matter much between 8 and 12 of the 1DII, the 1DIII and the 5D. The ability to practically see in the dark with a 50 1.2, is however, so attractive!

I might yet surprise myself.

Asher
 

John Sheehy

New member
Thanks, John, I got that from the outset. O.K., now what do we do with this? IOW, here do we go with this is a practical sense?

Look for ways to read out sensors with lots of pixels in such a way that the user has a choice between slow readout/write times and highest quality, or fast shooting with binning.

There could be cameras with small sensors and tiny pixel pitches that take DSLR lenses. I have no doubt that a 500mm f/4L could take full advantage of a pixel pitch of 2 microns, and a small-sensor like that in the FZ50, with more money invested in high-ISO read noise be greatly superior to using TCs on current DSLRs. TCs are made to magnify the focused depth well, but trash the bokeh.

What pictures in what industry or purpose would that Panasonic Camera then trounce the 1DII. This question is asked not as a confrontation. I believe that there must be a zone where the Z50 excels. But where is it in practical terms?

I did not compare the two cameras to suggest that the FZ50 "trounces" the 1DmkII in some practical way, in normal use. My point was that, as much as the 1DmkII is lauded for it's photon capturing ability, it does so by the brute force of sensor size, and not because of pixel size or efficiency. The FZ50, which is one of the cameras that makes people talk about how bad it is to cram pixels into a sensor, actually outperforms the 1DmkII at the square mm level in some ways, excluding read noise at high ISOs (the FZ50 has no high-ISO readout optimizations). The full-well photon capture of the two cameras, measured by square mm, instead of by the pixel, is almost exactly the same for both cameras. That's at ISO 50 for the Canon, and ISO 100 for the Panasonic. At ISO 100 (50 is missing some DR on the 1DmkII), the FZ50 collects 51% more photons at RAW saturation, per unit of area, and the read noise is smaller too (60%), when adjusted for spatial resolution. IOW, if you could find a way to replace the 1DmkII sensor with a bunch of FZ50 sensors, you would have *less* noise at ISO 100, and 4x the linear resolution.

A lot of people seem to that putting more and smaller pixels onto a given sensor size automatically and intrinsically decreases DR, and increases noise, but this is not necessarily true, and certainly isn't true in the range of DSLR pixel pitches.

It's a rut of false reasoning caused by the fact that smaller pixels have more noise at 100% view, and almost no one has seen a huge sensor with 2u pixels, to compare printed images at the same size, or monitor images with the same magnification of a unit of sensor area. That's why I decided to compare crops from DSLRs against crops or full frames of small-sensor cameras with the same focal length lens. Here are two examples (my biggest pixels, 7.4u from the 10D, and 3.4u from the Sony F707, and 1.97u from the FZ50):

10D on the left, F707 on the right. Same true focal length, same subject distance, same f-stop, same shutter speed, same ISO:
original.jpg



10D on the top, FZ50 on the bottom, same everything again, ISO is 1600 this time. Upper right and lower left are 100% pixel view; upper left is 10D upsampled to match subject size; lower right is FZ50 binned 3x3 (after demosaicing) to approximate 10D 100% magnification:
original.jpg



After doing this test, I found that not only is the FZ50 not optimized for ISO 1600, it is *unoptimized*; better FZ50 results would have been had at ISO 100 pushed 4 stops.
 

John Sheehy

New member
John,

Noise and DR per unit of sensor area? That's an unusual unit.

But the only one that matters, when you are viewing at a fixed amplification. Traditional noise statistics lie in the real world, where noise has width and height, in addition to the depth that is measured as standard deviation. Look at this graphic, both at normal monitor distance, and from across the room:
original.jpg



All 4 outer squares all have the same standard deviation, with two different types of noise (2-D and 1-D), and 3 different frequencies for the 2-D. The insets are downsamples of the outer squares; you can see that the noise has different strenghts at different frequencies, and reduces at different rates dowsampling. Noise is not as simple a thing as "standard deviation". You can only compare standard deviations directly when you are viewing at the same pixel magnification, and the noise is at the same frequency and in the same dimensions.

Perhaps what you're trying to say is that the 1DsII (or even the rumored 22MP 1DsIII) are not the end of the line for pixel density, as demonstrated by the FZ50? That's an interesting argument, if it's what you are in fact saying.

I don't think there are too many situations where the FZ50 will capture a better image than a 1DsII, but I suspect you're correct that there's yet mileage in the CMOS sensor beyond the 1DsIII.

All those amplifiers at the pixel site are going to be a problem with Canon technology.

However, when you get to smaller pixels, you don't *need* the components that digitize captures of 50,000 to 80,000 photons any more. The ISO 100- and 200- related components can be dropped, and the ISO 400 components become the ISO 100 components, if you quadruple the number of pixels, as you are reading a max signal at each photosite 1/4 what you were at the same ISO with 1/4 the pixels.
 

Ray West

New member
Hi John,

wrt the graphic, I think their are jpeg issues with the smaller squares, seems to be lack of contrast. Can you put a tif (maybe a gif) image somewhere.

Best wishes,

Ray
 

John Sheehy

New member
Hi John,

wrt the graphic, I think their are jpeg issues with the smaller squares, seems to be lack of contrast. Can you put a tif (maybe a gif) image somewhere.

I only saved it as JPEG. I toggled the preview checkmark on and off while saving, as I always do, and chose a quality of 10 out of 12 in PS CS.
 

John_Nevill

New member
So where does Fovenon sensors fit into the debate?, only kidding!

Anyhow, back on track with the original thread, UK shops have been taking deposits for while but have extended lead times. A colleague of mine is in the process of ordering but fears the "first batch syndrome" may deal him a dud, so he's holding off.

I sold off two bodies late last year and although tempted with the 1DmkIII, i'm thinking full frame and may will wait for the other replacements to come on line.

Although if Canon offer one up for review, i'll rise to the challenge!
 

Asher Kelman

OPF Owner/Editor-in-Chief
I think this discussion considering 2micron all the way up to 12Micron or more and the trade-offs in resolution, noise, DR and more merit a separate thread so these posts we made into the new thread to further explore this area.

One issue that I see holding back full exploitation of sensors is that the camera systems are closed to 3rd part firmware developers. We could see pixel binning and improvements in AWB in and optimizing for speed or DR for example and more if we were given or else hacked access!

Asher
 
Second, the sample image from FZ50 shows way more noise than the image taken by 1DII at a similar ISO level. You may be correct in theory but my customers would not be interested in one. They need a final image. And unless we're talking some artsy exceptions, noise=bad.

Which to me is the crux of the matter, the end result.

While John's (as I have called it before) provocative (which I like) view has some merit, I see some issues depending on practical use. To me, faithful rendering of the surface/material structure or 'fabric' of e.g. objects is the road to convincing/realistic images. One should be able and see/sense the difference between e.g. cardboard, leather, or varnish of the same color.

Since a single output pixel can only represent a single color, it is especially in the smooth gradients or uniformly lit areas of a certain color where low noise between pixels makes or breaks the realism of fabric. Afterall if a uniformly lit surface is of a single color, then all output pixels should render it the same within the physical boundaries of photon shot noise. That is where theory ends and practice rules.

The premise that if the individual output pixels are imaged small enough then they will visually blend into a lower noise average, fails when the images are relatively enlarged.

The pixel pitch of a display, broadly ranging from 0.22 to 0.29 millimetres is more than say 4 times worse than average human visual acuity, so non uniform pixels will introduce a visual impression of a bogus fabric. Even down-sampling of an image will introduce noise aliasing issues if the down-sampling is not properly executed (e.g. Photoshop's downsampling is not done as it should be done).

Printing of more noisy pixels on glossy material will need output resolutions in excess of 300 PPI, which will limit the maximum output size, especially for cropped images.

However, having said that, when the objects/subjects in the image have a material structure that has say 10x as much amplitude as the pixel noise, then the noise will not be as much of an issue and resolution becomes relatively important.

The real issue that remains is that with common (=affordable) technology, small pixels will have small potential well depths. That will lead to relatively higher photon shot noise and lower sensitivity. Both are unwelcome and the result will often distract from a superior viewing sensation (similar to e.g. a Large Format camera output).

Bart
 

Asher Kelman

OPF Owner/Editor-in-Chief
Bart,

i know this is a large request!

Still, could you make examples on the payoff or loss when photographing surfaces with detail of different amplitudes using ~2 microns v. ~ 7-9 micron or flat surfaces?

I'd like to see what the effect is on the screen and printing at 150, 200, 300 PPI resolution.

I hope I've framed the question corectly.

Asher
 
Bart,

i know this is a large request!

Still, could you make examples on the payoff or loss when photographing surfaces with detail of different amplitudes using ~2 microns v. ~ 7-9 micron or flat surfaces?

Totally flawless tests are hard to do, but I'll look around for some material samples with almost identical colors (in order to avoid major exposure and color balancing postprocessing effects which might bias the noise effects). I can also shoot some separate materials as they are, from different distances.

I can shoot that with my 1Ds2 (7.2 micron sensel pitch) and with a Powershot G3 (3 micron sensel pitch) at various ISO's, both with a DIGIC II processor in order to reduce variables a bit. There will probably still be some variation from the Raw processing, but we'll see.

I'd like to see what the effect is on the screen and printing at 150, 200, 300 PPI resolution.

I hope I've framed the question corectly.

Yes you've got it. I won't be able to do it this week, but I'll already start looking out for suitable stuff to shoot later.

We could also do some double blind testing on a single piece of material, asking for as many people as possible to cast a vote of preference for the most realistic fabric rendition in an unidentified random order pair of images (or pick the best of three from two cameras, or mix in different ISOs, to make it harder to guess). Let me ponder on a solid method to do that, suggestions from the forum members would be appreciated.

Bart
 

Asher Kelman

OPF Owner/Editor-in-Chief
Bart,

We can code the files, 3 copies of each. I'd get them printed with 3 different printers and then have people rank them as to realism with a scale of 1-5.

The sets would be rated by 40 people. 20 photographers and 20 non-photographers.

O.K. would that give us the numbers for answers.

Asher
 

John Sheehy

New member
Despite potential DOF issues, good one, keep them coming.
Solid statistical comparison methods are also appreciated.

The analog DOF will be the same for all, if the real focal length and f-stop are the same. CoC drops out of the equation when you are comparing resolutions, anyway. No value in counting resolution twice.
 

John Sheehy

New member
Yes you've got it. I won't be able to do it this week, but I'll already start looking out for suitable stuff to shoot later.

We could also do some double blind testing on a single piece of material, asking for as many people as possible to cast a vote of preference for the most realistic fabric rendition in an unidentified random order pair of images (or pick the best of three from two cameras, or mix in different ISOs, to make it harder to guess).

Why would you want to compare images at different ISOs?

My method of using the same Av, same Tv, same ISO, makes perfect sense. You are comparing what one sensor and readout do in the same focal plane area as another set.

I find it a bit humorous that you think that you can make it hard to guess; the fine-pixel-pitch camera will be blaringly obvious with its higher defintion, unless you bin or downsample it, which is actually a separate test.
 

Asher Kelman

OPF Owner/Editor-in-Chief
Why would you want to compare images at different ISOs?

My method of using the same Av, same Tv, same ISO, makes perfect sense. You are comparing what one sensor and readout do in the same focal plane area as another set.

I find it a bit humorous that you think that you can make it hard to guess; the fine-pixel-pitch camera will be blaringly obvious with its higher defintion, unless you bin or downsample it, which is actually a separate test.

Well,

Once the pictures are printed then what they came from will be hidden. In any case we are going to make all the files sizes the same, are we not? Is there a reason we shouldn't?

We could have one person bin or up-ress to standard sizes and then strip EXIF and delver 16BIT DNG files. Would that remove bias of the printers?

Asher
 

Ray West

New member
Is it possible for someone to define what is being proved here? What is the object of the exercise?

Best wishes,

Ray
 
I find it a bit humorous that you think that you can make it hard to guess; the fine-pixel-pitch camera will be blaringly obvious with its higher defintion, unless you bin or downsample it, which is actually a separate test.

Come on John, let's stop playing around. Unless I completely misunderstood you, your premise is that the (noisier) fine-sensel-pitch camera with binning/downsampling gives better image quality than a larger sensel of the same output resolution. That will require to match the binned output pixel size with the native one of the larger sensel camera, and to compensate for the optical differences because otherwise it would be too easy to spot which is which.

Nobody is trying to prove that inadequate/undersampled resolution can be compensated for by better dynamic range, if that's what you are thinking. We're exploring output pixel quality, and the effect it has on final material structure realism.

Bart
 
Is it possible for someone to define what is being proved here? What is the object of the exercise?

The issue at hand is multi-dimensional, which makes it a bit difficult to demonstrate unambiguously, there are too many factors at play. It is more involved than finding a balance between resolution and noise/DR.

There are roughly 2 visions on future sensor array developments, one states that the best image quality will be provided by more/smaller/noisier/hi-res sensels (with the option to bin for noise suppression), the other that the image quality benefits from more accurate/low-noise/hi-DR sensels.

The problem in demonstrating the benefit of either approach comes from the many technical differences between the two approaches, and how much of the differences can still be seen in the final output. I believe the ultimate proof is in the eating of the pudding, a high quality glossy print (which by itself introduces resampling variables, sigh).

Bart
 

John Sheehy

New member
Come on John, let's stop playing around.

I'm not playing around. There are two different kinds of comparisons possible, and I don't read minds. My main interest, and my main point, concerns the full resolution from both systems. Binning is something I mainly mention as a bridge concept, as it it probably easier for people to accept, even if they have been indoctrinated into a pixel-centric view.

Unless I completely misunderstood you, your premise is that the (noisier) fine-sensel-pitch camera with binning/downsampling gives better image quality than a larger sensel of the same output resolution.

I would call that "a" premise; it is not my main premise. My main premise is that resolution is more useful than low pixel-level noise, and that noise has height and width, as well as depth.

Nobody is trying to prove that inadequate/undersampled resolution can be compensated for by better dynamic range, if that's what you are thinking. We're exploring output pixel quality, and the effect it has on final material structure realism.

That's one test amongst others. The results of such a test should be fairly predictable, based on photon collection and binned read noise.

Such a test could be misleading; similar results could lead to the conclusion that there is no benefit to the smaller pixels, if you can not enjoy their increased resolution.
 

Ray West

New member
In principle, it would be easy to correct for most of the noise issues in software, but afaik it would not be software as we know it now. If that is done, then smaller pixels, will resolve smaller detail, until the pixels become all boundaries, if you get my meaning. However, for various reasons, it is not in any manufacturer's interest to get things right first time, but just make incremental improvements, and try and stay one jump ahead of the competition. So, what is theoretically possible will always be far ahead of what is available at any one time. The different results we can see, based on what is available, are a result of a number of compromises. At some stage, physics come into play, like electrons being to big to squeeze through the conductors, (if that is what electricity is, a flow of electrons???) or the length of the path being such that the delays become significant, or resonance occurs at certain frequencies.

fwiw, imho, I think what is being proposed re test shots is of little use, since there is unlikely to be sufficient control over the processing, etc. and the judging of results are too subjective. And, tomorrow, it will be different. Of course, such testing may well lead to plenty of discussion. ;-)

Best wishes,

Ray
 

Jack Joseph Jr

New member
Dogma Time. . .

"even if they have been indoctrinated into a pixel-centric view."

Good Lord, what kind of mumbo-jumbo is that? I'm always amused when someone who engages in mega-pixel-peeping or spec-reading tries to convince the rest us of something we know to be BS. Maybe my kids' Coolpix 2000s do some thing that my Canon equipment finds difficult, like DOF. So what? Keep your Panasonic Zwhatever. If it makes your life happier that's wonderful. I'll still use my inferior Canon bodies!
 
Top