The Missing Factor in Sony F828 Reviews

[Ray]

I am sure that believers in God are not foolish like atheists.  For a second, let's assume I am wrong - there is no God.  What harm have I done myself or anyone else by living the life of a believer?  But the long term harm for the atheist and the agnostic is unspeakable if they a wrong.  When you meet God, are you going to be provocative and deny Him?  Might as well.  It will too late to embrace Him.

Well, I'm certainly no judge of what harm you might have done yourself or anyone else by living the life of a believer, but I've read lots of stories of people who claim to believe in God, and yet appear from my perspective to do lots of harm to themselves and others. The Old Testament has many examples of believers doing great harm to each other.

[Ray]

In that case, what's all the fuss about? I don't have the camera to play with, but even Mattew Cromer seems concerned about CA.  He even returned the camera for an exchange, as have others, so I believe.

[Ray]

One suggestion at Norm Koren's site is that lens abberations starts to be the dominant factor in image quality at f/4 and wider, and he implies that this is an inherent limitation in optical design (dealing with coma, spherical abberation and all those other things that I know little about except their names).

Makes sense! There seems to be no hard and fast rule on this. I have some vague memory of a review of a Leica Summicron 50mm lens (or maybe it was a Carl Zeiss lens) that displayed its best performance at full aperture F2.8. I think there's at least one Canon lens that performs best at f4. However, the Canon 100-400 IS zoom, at 400mm, is sharpest at f11 which is exactly 2 stops down from full aperture.

I have some Sigma lenses which appear to be sharpest at about 4 or 4 1/2 stops down from full aperture   .

If we were to assume that the F828 Zeiss lens is sharpest at f5.6, then the fall-off at f8 is likely to be negligible, which would make the F828 an ideal camera for all those shots that require great depth of field.

No CA, great DoF and close to maximum sharpness. Why is no-one shouting about this?  

[Ray]

Who needs great (or greater) depth of field at focal lengths of 7 - 50 mm?

Who needs anything? What's your point? The pinhole camera is famous for its DoF, but the quality is lousy. I'd use F22 more often with my D60 if the resolution wasn't seriously compromised.

The resolution of the F828 seems to be on a par with that of the D10. But at what aperture? Resolution varies with aperture. Depth of field varies with aperture. Depth of field also varies with camera format, at any given aperture.

See the problem?

[BJL]

Probably DOF equivalencies are not quoted because there is considerable ambiguity about how one should compare. Two common options are (a) compare DOF as seen on prints of the same size and shape, viewed from the same distance, and ( compare DOF on the biggest prints that are consistent with the resolution that one expects from the sensor (including film emulsions as chemical sensors), again assuming the same minimum viewing distance, at least when details are being scriutinized.

   Case (a) leads to the guideline of having the same DOF with the same aperture diameter, or in other words, scale the f/stop in proportion to the chosen focal length: the lens of the 828 compares to a 28-200, f/8-f/11.

   However, case ( is perhaps more common in the film world, where people sometimes think in terms of the DOF requirements of prints at a maximum enlargement factor of about 8x (a 12"x8" limit for 35mm format and so on), so making larger prints from larger formats, and allowing for people sometimes viewing them from the same minimum distance despite the larger print size.

   At that extreme, the rule is that the DOF is dictated simply by the actual focal length and aperture, regardless of image format.

    With digital cameras, this comparison only makes easy sense if the pixel pitch is the same despite the different formats, but maybe a rough digital counterpart would be to work on the basis of printing at 300PPI and computing the DOF that one would then see when viewing from 10", which is about the distance that good human eyes need to be at to resolve 300PPI.

   I think that the guideline then becomes that one scales the f/stop in proportion to the pixel pitch (or to the resolution of the sensor/lens systems being compared, measured in lp/mm) when comparing DOF at maximum usable print size.

[BJL]

Depth of field is a function of f/stop, focus distance, and the acceptable circle of confusion size.  Period.  It has nothing to do with film/digital/format.  Again I suggest you read Michael's tutorial on Understanding Depth of Field.

Here we go again: format size DOES matter in DOF calculations because the circle of confusion size has to be adjusted for different format sizes: in the end, the CoC diameter that counts is the one on PRINTS after considerable enlargement from the CoC size formed in the camera, and images from a format like a 2/3" digicams with sensor diagonal of 11mm are typically going to be enlarged far more (about for times as much) than ones from a 35mm format sensor (diagonal 43mm), so the circle of confusion in the image formed in the camera needs to be far smaller (about 1/4 as much) than that used in DOF computations for 35mm film.

   One hint of this is to look at the CoC values chosen for various formats in a DOF calculator like the one at
http://dfleming.ameranet.com/dofjs.html
Change the format choice and see how the recommended CoC diameter adjusts.

    As another exercise, try changing from 35mm format, focal length f=80mm, aperture ratio f/16 to Sony 828, f=20mm (roughly the same angular FOV), f/4 (same aperture diameter of 5mm in each case), keeping the same focal distance. You will see that the near and far focus distances are about the same, whereas if you keep the f/stop at f/4 in each case, they are wildly different.

[Ray]

Next exercise; read my previous posting.  

[Howard Smith]

First, depth of field and format have no relationship.  The lens simply has no way of knowng what's behind it.  Depth of fieldhas nothing to do with resolution.

Yes, depth of field calculators do use different size circles of confusin for different formats.  You should read the fine print to understand why.  The reason is smaller faromats are usually enlarged more than larger formats.  An 8x10 contact is much easier to see than a 35 mm contact.  The calculators assume the user is using a lens of appopriate focal length.  That is why a "normal" 35 mm lens is about 50 mm, a "normal" 6x6 lens is about 80 mm, and a n
"noral 4x5 lens is about 150 mm.  Put a 150 mm lens on a 35 mm camera and you will get the same image size as a 4x5, just cropped rather severely.  Set bot lens at f/whatever, and you get the same depth of field when making a print with the same image size.

It appear that non of us re going to give up our positions, so let's just agree to disagree and go on.

[BJL]

Ray,

    I generally agree about comparing performance mostly with comparable angular field of view, depth of field and shutter speed, to most closely match the overall "composition", but with two small disclaimers.

a) Michael's assessment of the 828 might have been primarily aimed at deciding whether it is "good enough" for use as a more compact, lightweight alternative to the 1Ds kit in some situations, rather that a head-on "which is better?" comparison to the 10D; his testing seem fine for establishing that he can forgo the bulkier intermediate option of the 10D in many situations.

 As to the different question of "which is better?", there is also a place for comparing at the apertures determined to give best overal image quality; maybe f/5.6 is a decent first guess for many lenses, due to abberation problems that arise at apertures of f/4 and larger, more or less independent of focal length and format. I am sceptical of the dogmatic "two stops down from maximum", and note that Norm Koren makes the vaguer statement "2 to 4 stops" in his ineresting discussion at http://www.normankoren.com/Tutorials/MTF6.html#Sweet_spot

   But this is rather speculative. I would be very interested to see tests of abberation effects on image quality in various lenses at their larger apertures, since optical theory is so much more complicated and less illuminating for abberation than the formulas for circle of confusion and diffraction spot size (Airy disc diameter.)

[BJL]

The approaches suggested by Ray and Howard both make sense to me, according to which aspect of camera performance one is assessing. How about distinguishing the following cases?

a) No shortage of light, assessing ultimate image quality (like tripod shots with no moving subjects, or hand-holding in bright enough light): set each camera to its optimum ISO rating (usually the lowest) and each lens to its optimal aperture: let the shutter speed fall where it may.

 Light availability limited by shutter speed limits (due to subject motion and/or hand-holding) and also the need for adequate DoF: exposure time and aperture diameter are limited to about the same maximum, so adjust ISO speed setting as needed.

c) Light availability limited by shutter speed limits as above, but with no particular DoF needs: test at the same shutter speed, with each lens at its widest usable aperture, and again adjust ISO speed setting as needed. ("Widest usable aperture" might exclude a lens's maximum aperture if that leads to problems like the dreaded purple fringing.)

P. S. want to redo this whole debate with the next 8MP 2/3" format digicam, the Nikon 8700 with is 8.9-72mm [35-280 equiv.], f/2.8-4.2 lens?

[Ray]

Call me old fashiond (I am)

Well, you've said it.    I wouldn't strenuously disagree with you.  

The paradigm is changing and I don't think any of us really knows what the future will bring. But there seems to be a trend towards the smaller format equalling the quality of larger format, with digital.

The 1Ds, for all practical considerations as far as Michael is concerned, equals the image quality of the Pentax 6x7, and the F828 more or less equals the quality of the 10D (but not feature wise), with a bit of help from noise reducing programs.

I find it difficult to get my mind around the significance of the competing issues and trade-offs. How much can sheer numbers of pixels compensate for lack of size? How good can these short focal length lenses really be at f2 or f2.8? What are the relative economics of producing a smaller format with the same image quality as the (usually) more expensive larger format?

These are questions to be considered.

[Ray]

Unless quantum mechanics has changed a great deal since I was in college, I think I can understand and appreciate the subtle difference betwee digital and analog.

Howard,
I believe you understand the diffefrence between digital and analog, and I think I more or less said so in my previous post. Sorry if you've misinterpreted my comments.

Nevertheless, you do not seem to have understood the significance of a truly digital camera. All the current debate on noise, as it applies to small format cameras in relation to large format cameras, is essentially an analog problem. We don't actually have any digital cameras. They're all analog cameras with 'attendant' digital functions that result in a digital output.

As Moore's law continues to apply, and I'm assuming here that it will, 10GB images will eventually present no processing problem. When this situation arrives, the stage will be set for a true 'digital' camera with no discernible noise, and a dynamic range of 20 f stops. The large format (and medium format) camera then becomes a total irrelevance; no more than a museum piece. That's the point I was trying to make.

Okay! Still friends?

[Howard Smith]

BJL, no, I have no proof or evidence.  I should have stated that as opinion based mostly on Canon finally making a full frame 35mm.  That camera seemed to eagerly awaited and overdue.  Canon may have had motvation to make a digital body that could use their rather complete line of 35mm lenses to best advantage, rather than introduce all new lenses.  But then Nikon should be similarly motivated - both to keep up and use their existinglenses for now.

Because Nikon and Canon arre not currently players in medium or large format, I wouldn't expect then to introduce a new camers line.  Of course, I was surprised when Hasselblad brought out the X-Pan line.  I would expect that Hasselblad will make an effort to have a digital back for there current medium format camera line.  They have always prtty good at keping the old 500 alive.

[BJL]

BJL, no, I have no proof or evidence.  I should have stated that as opinion based mostly on Canon finally making a full frame 35mm.  

Perhaps this is mostly a confusion over tense, past versus present. Clearly there has was for a long time a need to develop larger sensors, because to some degree greater size has advantages. (However, my crude analysis of how noise levels vary with sensor size, aperture and exposure time suggests that the advantages hold under more limited circumstances than many believe, requiring adequate light levels and/or decreased depth of field if resolution and/or dynamic range is to be improved over what smaller sensors can achieve. The failure of the three "full frame" DSLR's to show better overall low light performance than the APS models with sensors less than half as big corroborates this.)

   I was talking of the present situation, where sensor size growth seems to have mostly or completely stopped (especially with digicams, where it stopped years ago). Since you mention Canon, they are an interesting case; the 1D Mark II does not, as widely predicted, move up to full 35mm format but instead stays at the same size as the 1D (or is there a very slight increase?); the amateur level sequence of D30/D60/10D/300D has shown no sign of increasing sensor size, but instead just one decrease in photosite size, accompanied by an improvement in noise levels.

[Howard Smith]

Ray, I was not aware that large format lenses have superior resolution than 35mm lenses.  In fact, I have always thought just the opposite.  And I also thought that the resolving capability of, say, T-Max 100 was the same whether it is 35mm, 6x6 or 4x5.

I always thought that the advantage of large format was not its lens, but the fact that a great deal of the enlarging is done at exposure time, so the negative need not be enlarged as much to make a print.

[Howard Smith]

OK, I think I have it.

If you can make a small sensor digital camera that is better than a current large format film camera and don't bother to improve the current large format camera, you will have a small format digital camera that is better than the current large format cameras.

That makes perfect sense to me, and was so obvious that it is hardly worth mentioning in the first place.

[Lin Evans]

No CA, great DoF and close to maximum sharpness. Why is no-one shouting about this?  

They really should be. The thread has strayed a bit from the original question, but the F828 is exceedingly sharp at F8. I shoot gallery art for a living, and there are times when I'm using one of my several dSLR's or my Kodak MF back when I simply long for the incredible DOF offered by the small sensor digicams which capture DOF and sharpness with ease which is downright difficult to achieve with my conventional film or digital SLR's.

The F828 has offered the first truly useful compromise which I can actually use professionally for those times when I really don't want to spend all day trying to achieve the DOF and "look" I want from my images. Because of it's increased resolution over the present digicam field (Nikon's offering not yet tested), I find I can get quite suitable 16x24 prints with near perfect color match and incredble sharpness, especially when doing relatively close-up work.

Here's a quick sample of a Jon Anderson original polymer art piece (about 3 inches long) at reduced size (800x600) and a link to the full sized, slightly cropped version. ISO 64, F8, full tele (51mm true - 200mm equiv) 1/400th

Lin



http://www.lin-evans.com/f828/horntoadb.jpg

[Ray]

The worse S/N ratio and dynamic range of more numerous smaller individual pixels is exactly balanced by the greater number of them used to print each given portion of the final image.

BJL,
You've made this concept very obscure by setting so many convoluted conditions.  

As I understand it, there are three broad types of noise; photon noise; dark noise and read noise. It's the additional read noise (the process of quantifying the electronic signal) which might be unavoidable if one substitutes fewer bigger pixels for a lot of smaller ones.

For example, let's suppose that Canon were to replace the 10D with a 12MP model, keeping the sensor the same size, the design principles and quality of components the same and the noise reduction processes the same etc., then the total read noise in the entire image would have essentially doubled, would it not?

[BJL]

BJL,
You've made this concept very obscure by setting so many convoluted conditions.

My apologies: which conditions were convoluted? There were really only three:
a) Comparing sensors of the same size and technology (e.g. Sony's 2/3"  CCD's at 5MP and 8MP)
 Comparing images of same size rather than the widespread nonsense of comparing files of different MP counts at 100% pixel size on screen
c) not going to photosites so small that a larger fraction of the sensor surface is occupied by stuff other than electron wells, reducing highlight headroom.

   Only the last seems convoluted, but it simply recognizes one of the practical limits on photosite miniaturization.

[Ray]

Can you give me a source for that "Read Noise figures as low as 5 to 11 electrons for CCDs with a full well capacity of 45,000 to 375,000 electrons."?  Because if it means "5 for 45,000 and 11 for 375,000", then there is a considerable reduction of read noise with pixel size, very close to my beloved square root pattern.

BJL,
Actually, it's the other way round. The larger read-noise figure relates to the smaller photodetector. This appears to be due to the frequency of the read rate in the two examples chosen. The higher the sampling frequency, the greater the read noise.

I can find no information on Roper's site that expresses a relationship between physical pixel size and read noise. The two examples I quoted can be found at:-

http://www.roperscientific.com/library_enc_dynamic.shtml

[/QUOTE](Do you really believe that CCD's are old hat and that most or all present progress is in the CMOS world? If so, explain it to all the digital back makers, and to Fuji, whose SuperCCD is the high resolution, low noise measurement leader in the APS DSLR field.)

Hey! What do I[/i] know!    I'm just an ordinary guy with a few scraps of information and a few half-baked ideas trying to fill in the enormous gaps.  

CMOS just seems better positioned to take advantage of the rapid development in nanotechnology and chip manufacturing processes used by companies such as Intel. As it becomes possible to manufacture increasingly smaller components, you could end up having a whole 'city' of operations within that one 4 micron photosite.