sharpness, megapixels, sensor sizes

[mikesnorth]

Hi everyone

I have recently discovered an intriguing "phenomenon" that has confused me totally...  

We tested a P25 with H1 body and 50-110 lens versus a H4D50 with same lens.
The P25 produces sharper images at anything greater than f/16.

I spoke with the person who put me on to this and he says he has tried this on many MF digital backs from 22mp to 50mp, (Phase One and Hasselblad) and always finds the same results.

My first thought was software. Capture One by default applying a greater sharpening mask than Phocus.
However, the P25 is sharper than other, larger, Phase One backs so this can't be the case as the software is of course, the same.

This is not a Phase One versus Hasselblad debate rather an enquiry as to possible reasons for this sharpness issue...
Sensor size vs number of pixels? Pixel density?

I found that you can achieve level of "sharpness" in Phocus with a H4D50 by increasing sharpness threshold to 350. Doesn't explain difference within same brand backs using same processing software though.  

Any thoughts/suggestions would be appreciated.
Thanks

[BJL]

I have recently discovered an intriguing "phenomenon" that has confused me totally...  

We tested a P25 with H1 body and 50-110 lens versus a H4D50 with same lens.
The P25 produces sharper images at anything greater than f/16.

How are you comparing sharpness?
1. Viewing equal sized prints?
2. On-screen viewing of the same part of the image?
3. Viewing on-screen at 100% (or equal percentage), so that you are viewing a smaller portion of the H4D50 image at greater magnification?
4. Other?

I would expect 3. to give the results you describe, since the greater enlargement would make diffraction effects more visible at those high f-stops.

[georgl]

Yes, the "pixel size" (or to be more precise: the "pixel-pitch", the distance from one center of a photosite to the next) is different. The 22MP-MFDBs like the P25 incorporate a 9µm pixel-pitch, while the 50MP-backs use a newer generation of sensors with only 6µm pixel-pitch. Don't mix it up with the actual size/sensitive area of the photosites (the older 9µm-sensors have larger non-sensitive "gaps" between individual photosites) and the effects on noise/DR, but it makes the process more demanding. While you need "only" 55lp/mm to reach the sensor limit on the P25, you need 83lp/mm to reach the nyquist limit for the H4D50.
Of course the lens resolves less contrast at 83lp/mm than at 55lp/mm, diffractions becomes an issue (f16 should already decreae contrast around 83lp/mm noticeably) aberrations become more visible (it's like a stronger enlargement) and the final image therefore appears softer when seeing it at 100% (it's like making two prints from the same file, one with 40x60cm and one with 60x90cm (50% linear increase, just like 6->9µm pixel-pitch), watching the prints at the same distance makes the larger print appear softer).
But of course the actual resolution within the whole image is still actually higher than with the 22MP-image. To resolve similar sharpness at 100% you need a careful workflow (focusing, vibration...) and a better lens.

[fredjeang]

Yes, the "pixel size" (or to be more precise: the "pixel-pitch", the distance from one center of a photosite to the next) is different. The 22MP-MFDBs like the P25 incorporate a 9µm pixel-pitch, while the 50MP-backs use a newer generation of sensors with only 6µm pixel-pitch. Don't mix it up with the actual size/sensitive area of the photosites (the older 9µm-sensors have larger non-sensitive "gaps" between individual photosites) and the effects on noise/DR, but it makes the process more demanding. While you need "only" 55lp/mm to reach the sensor limit on the P25, you need 83lp/mm to reach the nyquist limit for the H4D50.
Of course the lens resolves less contrast at 83lp/mm than at 55lp/mm, aberrations become more visible (it's like a stronger enlargement) and the final image therefore appears softer when seeing it at 100% (it's like making two prints from the same file, one with 40x60cm and one with 60x90cm (50% linear increase, just like 6->9µm pixel-pitch), watching the prints at the same distance makes the larger print appear softer).
But of course the actual resolution within the whole image is still actually higher than with the 22MP-image. To resolve similar sharpness at 100% you need a careful workflow (focusing, vibration...) and a better lens.

I think that it is indeed important.

In the testings I've been able to acheive to know what I really need (with C1 but it has little to do with the software itself IMO), the greater are the resolution and pixel-pitch are indeed more demanding and all the chain involved has to be on par.

I found (today) that the balanced between PP workflow-price and IQ is "desequilibrate" with the biggest backs because one little "mistake" on the chain like lens quality means that you do not reach the expected results.
In that sense, I do agree 100% with Michael Reichmann, who has reiterate this fact over and over again here.

I can not buy an high-end back, but in the personal testings have been doing with rented or files nicely provided by friends etc...that even if I had the money now to invest in such backs, I would not go further than 40MP. In fact, 20-30 is the ideal compromise IMO. (I'm not saying here that the high-end backs are not superiors)

Keep in mind that the printings, when made by real gurus (and in that sense I doubt that the photographer can be both the artist and the printer to the same level, but that is another debate), can do miracles with 22MP files. I am talking about really big prints. Not a long time ago, when 60MP backs did not exist yet, some artists where doing big enlargment with the availble resolutions and it was great, and they were sharp.

[mikesnorth]

Thanks georgl

Been researching causes, etc all morning and what you said regarding pitch is precisely what i managed to conclude as well.

So, reduction in pixel size leads to less noise effectively.
And reducing pixel size leads to more pixels on the sensor thus more information and detail being recorded in terms of tonal range, etc. ?

So do you basically have to sacrifice some sharpness for more detail and a larger image?

Thanks, really appreciate all the input



[quote name='georgl' date='Jun 4 2010, 11:55 AM' post='369434']
Yes, the "pixel size" (or to be more precise: the "pixel-pitch", the distance from one center of a photosite to the next) is different. The 22MP-MFDBs like the P25 incorporate a 9µm pixel-pitch, while the 50MP-backs use a newer generation of sensors with only 6µm pixel-pitch. Don't mix it up with the actual size/sensitive area of the photosites (the older 9µm-sensors have larger non-sensitive "gaps" between individual photosites) and the effects on noise/DR, but it makes the process more demanding. While you need "only" 55lp/mm to reach the sensor limit on the P25, you need 83lp/mm to reach the nyquist limit for the H4D50.

[mikesnorth]

This is exactly how we compared.

Why are you viewing at greater magnification? Is it not the same for both images? Of course the portion of the image viewed is bigger with the bigger file but isn't viewing at 100% an equal way as it is full size for both images?  

Thanks for the input. really appreciate it!!


3. Viewing on-screen at 100% (or equal percentage), so that you are viewing a smaller portion of the H4D50 image at greater magnification?

I would expect 3. to give the results you describe, since the greater enlargement would make diffraction effects more visible at those high f-stops.
[/quote]

[ErikKaffehr]

Hi,

Are you sure you have correct focus on the H4D50? Best sharpness should be apparent at about f/8, so if you need to stop down to f/16 something is wrong.

Check this article: http://www.josephholmes.com/news-medformatprecision.html

Best regards
Erik

How are you comparing sharpness?
1. Viewing equal sized prints?
2. On-screen viewing of the same part of the image?
3. Viewing on-screen at 100% (or equal percentage), so that you are viewing a smaller portion of the H4D50 image at greater magnification?
4. Other?

I would expect 3. to give the results you describe, since the greater enlargement would make diffraction effects more visible at those high f-stops.

[BJL]

This is exactly how we compared.

Why are you viewing at greater magnification? Is it not the same for both images? Of course the portion of the image viewed is bigger with the bigger file but isn't viewing at 100% an equal way as it is full size for both images?  

Thanks for the input. really appreciate it!!


3. Viewing on-screen at 100% (or equal percentage), so that you are viewing a smaller portion of the H4D50 image at greater magnification?

I would expect 3. to give the results you describe, since the greater enlargement would make diffraction effects more visible at those high f-stops.

Forget pixels for a moment, the issue is magnification from the image formed by the lens at the focal plane to the image displayed. If the image viewed (on screen or on paper) of the same subject from one camera is larger than that from the other, then there is greater magnification, including great magnification of the effects of lens aberrations, diffraction, out of focus effects, camera motion, subject motion, and so on. Indeed, you will see the same decline in sharpness when you compare different sized prints from the same film or negative.

Since you said that the effects are most noticeable at high f-stops, beyond f/16, the likely explanation in this case is that at f/16, the diffraction spots are considerably more than the 6 microns size of the photosites on the sensor of the H4D50, so diffraction will be visible at 100% pixels on screen. On the other hand, the diffraction spots at f/16 are barely large enough to be resolved by the 9 micron pixels of the other sensor, and will be less visible or invisible with the smaller displayed image size from that camera.

Other replies have mentioned other possible factors, but given the connection with high f-stops, it seems that diffraction is the main cause in your case. For the record, though:
if you wish to exploit a sensor (or film) of higher resolution to display images at larger sizes or view them more closely, you have to deal with issues like

1. Limiting aperture choices to a somewhat lower maximum f-stop in order to reduce diffraction effects: a common guideline is f-stop not more than twice the pixel spacing in order to get full sensor resolution, so about f/11 or less for the H4D50.

2. Limiting aperture to a somewhat higher f-stop with a given subject in order to reduce out of focus effects on parts of the scene that you wish to be sharp. (That is, getting adequate DOF gets tougher when images are displayed at a greater degree of enlargement.)

3. Better stabilization of the camera.

4. Higher shutter speeds to reduce effects of subject and camera motion.

5. Sharper lenses.


And yes, items 1 and 2 represent an inherent conflict when you want a very large, sharp, detailed image of a subject that does not lie entirely in or close to a single plane on which exact focus can be achieved! Stitching can help with lens aberrations by allowing the use of a lens over a narrower angular range (like using a normal or short telephoto lens in a smaller format and stitching vs  using a wide-angle lens in a larger format), but beyond a certain level of desired sharpness on a given subject, only techniques like focus stacking can help.

[mikesnorth]

Thanks, getting clearer now...kinda  

I guess my main concern then, assuming all this can be explained, is why the camera manufacturers can't compensate for this. They of course know about it. Is it not strange that it occurs?

And then, how does a person obtain equally sharp images using a 50mp camera for example?

The "tests" we did were with studio flash and shot at 1/350, surely enough to handle camera shake, etc?

[Jason Denning]

So what would be the best megapixels to use f22 on a lens? 22?

[quote name='BJL' date='Jun 4 2010, 08:05 AM' post='369450']


1. Limiting aperture choices to a somewhat lower maximum f-stop in order to reduce diffraction effects: a common guideline is f-stop not more than twice the pixel spacing in order to get full sensor resolution, so about f/11 or less for the H4D50.

[Doug Peterson]

As others have said but I hope to condense and reiterate:
1. Diffraction (as others have stated) is clearly limiting the resolution of the smaller-pixel-size camera
2. Comparing two images of different resolution at 100% on-screen is a comparison of sharpness rather than detail. A print of equal size would compare total detail. Both are useful measures but are not the same thing.

You may find that different raw processors handle micro detail better (or at least differently). Take the same raw file and run it through Capture One, Irrident Raw Developer, Aperture/iPhoto/Phocus, LightRoom and Bibble and you will see different micro-detail results (regardless of the sharpening settings). However, that difference is minor compared to the diffraction you are experiencing. So if using higher aperture values is important to your work you should experiment with different lenses you own or could rent/buy to see which will perform best for you.

Finally, lens design is a consideration when examining diffraction. Take for instance the link and excerpt below showing the comparison we did using the same digital back but with a Phase One 120mm D focal-plane macro lens and a Hasselblad HC120mm leaf-shutter macro lens: http://www.captureintegration.com/tests/phase-one/

I guess my main concern then, assuming all this can be explained, is why the camera manufacturers can't compensate for this. They of course know about it. Is it not strange that it occurs?

Diffraction is physics. It's like asking why we can't go faster than the speed-of-light or asking why a car with a gasoline engine built for fast-acceleration/top-speed has poor gas mileage.

Doug Peterson
__________________
Head of Technical Services, Capture Integration
Phase One, Leaf, Cambo, Canon, Apple, Profoto, Eizo & More
National: 877.217.9870  |  Cell: 740.707.2183
Newsletter: Read Latest or Sign Up
RSS Feed: Subscribe
Buy Capture One at 10% off
Personal Work

[bcooter]

Diffraction is physics. It's like asking why we can't go faster than the speed-of-light or asking why a car with a gasoline engine built for fast-acceleration/top-speed has poor gas mileage.

I look at these tests and my brain kinda fries and  before I ever look I know the manufacturer's/dealer's  samples will always look better than the competitors.

Still, this is just stuff to me, not real world working and maybe the next time you do this throw an old Contax into the mix.

This 100% crop was shot with a 120mm contax macro F16 and I don't think I need to ever see more detail.  Actually I'm not sure if more detail would ever reproduce anyway.

[attachment=22404:contax_f16.jpg]

Then moving on to real world working, this was with the Canon 1ds3 at F 13 which I guess by your calculations of the poor old dslrs cramming too many pixels into too small a space would show defraction, noise, taco juice on the model's lips,or something awful that according to all medium format dealers,  all dslrs do.  

[attachment=22405:canon_f13.jpg]

Anyway, Maybe there could be more detail but I doubt if it would look more film like or give anyone anything else they want to see.

Now from the same session is real world changes.  This was shot at 400 iso at F 3.5 with modeling lights.  (This is just a screen shot from bridge).  

[attachment=22406:canon_400iso.jpg]

Rather than rework all the lighting, put on ND's to lower the output to throw the foreground and background material out of focus, all I did was just turn off the pocket wizard, open up to 3.5, move the iso to 400 and shoot.  It's tight, it's more than useable, it more than enlarges to almost any size, so back to real world, time is a premium and the ability to move this camera around is something that is hard to explain unless your on set.  If I tried this with a p30+, I'd still be cloning and smoothing out noise.

And before you mention sensor plus, I just don't get 10 mpx for $20,000.    If I need high iso there are Canons out there that go to a trillion iso, for  $2,800 that shoot beautiful 22mpx stills, high def video, tether, have hdmi ports and live view and nobody in the professional world goes to work with just one camera anyway.




BC

[mtomalty]

Diffraction is physics. It's like asking why we can't go faster than the speed-of-light or asking why a car with a gasoline engine built for fast-acceleration/top-speed has poor gas mileage.

....and, I suppose, also explains why can't Phase source a great LCD   :>))


Mark

[ErikKaffehr]

BC,

I don't disagree with your writing, except:

1) There is now way you achieve optimal resolution on a small pitch (say 6 my) camera at f/16.
2) If shelling out 30 kUSD on a 60 MP back you probably don't want to turn it into a 30 MP back by using small apertures.
3) In case all those pixels are really needed, the only way to achieve them is exacting work.

A few comments:

In some situations we can quadruple resolution by stitching. Stitching effectively increases sensor size. Obviously not the way to go with the kind of commercial work you do.

Sharpening works well with diffraction effects. The lost resolution is not regained but it can compensate nicely for the reduction in edge contrast.

This article by Joseph Holmes indicates that there are a lot of problems with rental equipments in MF: http://www.josephholmes.com/news-medformatprecision.html
Still, plenty of photographers use rental equipment without complaining.

My guess is that the need for megapixels is often exaggerated, and hard to achieve when really needed.

The figure below is from an experiment I made using APS-C cameras with different pixel pitch, 8 and 6 microns. There may be experimental errors but the effect of diffraction is very clear at f/16:

http://echophoto.dnsalias.net/ekr/images/s...actionlimit.gif

Best regards
Erik

I look at these tests and my brain kinda fries and  before I ever look I know the manufacturer's/dealer's  samples will always look better than the competitors.

Still, this is just stuff to me, not real world working and maybe the next time you do this throw an old Contax into the mix.

This 100% crop was shot with a 120mm contax macro F16 and I don't think I need to ever see more detail.  Actually I'm not sure if more detail would ever reproduce anyway.

[attachment=22404:contax_f16.jpg]


BC

[Steve Hendrix]

In every case I've seen, 22MP DB's with 9 micron sites, 60MP DB's with 6 micron sites, Contax Zeiss lenses, Hasselblad HC lenses, Phase One/Mamiya D lenses, Rodenstock HR's and Schneider APO's, f16 to f22 is the great divide: The point at which apparent sharpness takes the big hit. Naturally these lenses are typically sharp at f8-f11 (and some even more so at wider apertures) and for my money f16 usually is very usable. But f22 is the point at which - depending on the lens and the digital back - the apparent sharpness really comes into question for those who place a high premium on sharpness. Beauty and fashion perhaps not so much. But for some markets and some photographers it matters. Last week I had a photographer tell me a client told him his images are sharper than his competitor and it mattered to the client. He does not shoot beauty or fashion or - portraits for that matter.

As far as Sensor Plus, since it was brought up, I think 10MP is a push (though a huge amount of images in the world are captured by 12MP DSLR cameras). I like 15MP at high ISO from a high quality digital back in Sensor Plus mode over a non-cropped, large sensor. But I don't really see Sensor Plus as intending to replace DSLR for high ISO (though it has for some). I do see it as a tool for someone shooting medium format to be able to shoot medium format in more situations or to continue shooting medium format in changing situations, rather than switching cameras, lenses, ratios, as a lighting situation changes. Not a miracle worker, but no question a beneficial tool and most of our customers shooting P40+/P65+ are surprised to realize they use it more than they thought they would.

I think also that Sensor Plus was a very important development for Phase One because if the next generation of sensor is indeed 90MP or more (as I expect) having Sensor Plus mode will be essential. Being limited to having to capture 90-100 megapixels with each shot will severely limit the market for such a product, I believe. When you can push a button and shoot 22-25 megapixels and gain 2 stops in the process while still covering a large sensor area 645 or larger, then you have something close to what many have pointed to for years as the right balance in sensor dimension/resolution.


Steve Hendrix

[ondebanks]

f16 to f22 is the great divide: The point at which apparent sharpness takes the big hit. Naturally these lenses are typically sharp at f8-f11 (and some even more so at wider apertures) and for my money f16 usually is very usable. But f22 is the point at which - depending on the lens and the digital back - the apparent sharpness really comes into question for those who place a high premium on sharpness.

Steve, exactly. So, what to do if the situation demands f22? Well - as Doug said, and as I teach - you can't beat physics. So if you must shoot at f22, the only way to achieve higher net image quality is ... to stop worrying about pixel sizes (anything from 12 microns down is fine), and just use a larger area sensor. More mm^2. Something that the large format film guys have known for decades.

[fredjeang]

I think also that Sensor Plus was a very important development for Phase One because if the next generation of sensor is indeed 90MP or more (as I expect) having Sensor Plus mode will be essential. Being limited to having to capture 90-100 megapixels with each shot will severely limit the market for such a product, I believe. When you can push a button and shoot 22-25 megapixels and gain 2 stops in the process while still covering a large sensor area 645 or larger, then you have something close to what many have pointed to for years as the right balance in sensor dimension/resolution.


Steve Hendrix

I guess that very soon the biggest film view cameras will not make any sense except the price, because I also guess that the next high end backs generation will also put, not the house, but the castle mortgage in trouble.

Anyway, I'm asking myself if these resolutions are going to be suitable for any alive being image because is it not going to be a kind of dermathologist imagery? We will see the retouching task directed to kill even more the "exessive" details in order to make the model look human and attractive and not a skin-medical-research image. Yes, for arquitecture and lands that is great because the maniac of details.

If you are going this way, we will enter directly into the textile fibers and microbians life one day and we will need to watch the pics from at least 50m distance.

Funny, some years ago retouching was looking to enhance details, tomorrow they will try to get rid of them.

[joofa]

It's like asking why we can't go faster than the speed-of-light ....

I thought there was that hypothetical particle Tachyon that is supposedly faster than speed of light! However, just like many aspects of modern physics today that seem more science fiction than science, it is not yet "discovered"  .

Joofa.

[fredjeang]

I thought there was that hypothetical particle Tachyon that is supposedly faster than speed of light! However, just like many aspects of modern physics today that seem more science fiction than science, it is not yet "discovered"  .

Joofa.

With your permission,

I thought there was that hypothetical MF back that is supposedly faster than speed of tablets!
However, just like many aspects of modern MF today that seems more DIY than industrial, a proper LCD is not yet discovered !

[bjanes]

As others have said but I hope to condense and reiterate:
1. Diffraction (as others have stated) is clearly limiting the resolution of the smaller-pixel-size camera

Actually, diffraction has nothing to do with pixel size and it produces the same effects in the image plane for both large and small pixels and also with film. If the image is diffraction limited with large pixels, using smaller pixels will produce the same or better resolution. The oversampling with the smaller pixels will reduce phase effects, aliasing, and moire. In addition, the smaller pixels will allow greater resolution at wider apertures.

Diffraction is physics. It's like asking why we can't go faster than the speed-of-light or asking why a car with a gasoline engine built for fast-acceleration/top-speed has poor gas mileage.

In the previously quoted article by Joseph Holmes, Mr. Holmes (who is a pretty serious guy and careful photographer) points out that the effects of diffraction can be mitigated through the use of deconvolution (see Optical Issues, section 3). He appears to be using Smart Sharpen in CS3, and I doubt that the CS3 algorithm is sufficiently sophisticated to accomplish this task, since it uses a very simple point spread function (PSP) for the deconvolution. Thus far, deconvolution has been limited by the lack of a suitable PSP. In the future, perhaps blind deconvolution and other advanced techniques will produce better results.