Top-end digital concerns

[bglick]

> OK, so I reviewed this - and I understand some of the logic you use, but I have to admit it makes no sense to me in real life.


It will soon.... remember, the f #'s and their MP's are MAX. POSSIBLE recorded resolutions, not min. resolutions. This also represents the resolution at the POF (Point of exact Focus). So this assumes you are shooting a FLAT test target or an infinity scene, both having a flat focal plane. When you introduce depth into the equation, the near and far points will always have less resolution than the POF, (up to a point) so the f# and MP's values I offered above, you can cut them by 20 - 50% (estimate) based on the amount of depth you have in a scene. This is why the 568 MP values are so far from reality.... real world depth MP values for 8x10 at f45 can be in the 40 - 90 MP range, based on the amount of depth and f stop used. (more on this below) There has been several web examples of 40 MP digital captures nearly matching 8x10 film - same composure being compared...(color, not B&W) .... and when there is enough depth in the scene the gap can be surprisingly small...but shoot infinity subjects or FLAT subjects, and 8x10 film will destroy a 40 or 60 MP back. This is why the comparisons are all over the map... the devil is always in the details....


Also, as per my f# vs. MP values above, ..... in the real world, nothing is perfect, so these values are not fully attainable even with FLAT subjects such as test targets....but with enough precision, the right lenses, used at the right f stops, optimum subject contrast, optimum subject sizes, precise processing, etc., in theory they are attainable ....


Film records contrast....the finer the subject detail, the less contrast the lens will deliver, and the less detail the film can record. This is why you must combine the MTF of the lens and the film, and why they are both dependent on image frequency (amount of detail) Just look at the MTF curve of a lens, at high enough resolutions, their ability to deliver clean contrast to the film, goes right in the dumper. Bruce mentioned a perfect example, pine needles at a distance.... low contrast and very fine detail, in most cases, 1/R will turn these needles to mush on film... sad, but it's reality. It's the nature of imaging, not just film, same true with digital. But digital has the benefit of lower f stops, due to more dense recording media. This is films Achilles heel. Anyway, this is why there is some generalizations when making these "ballpark" judgements. In most cases, its to get us in the ballpark... unless you really want to get down to the sheer nitty gritty stuff...



> I recently did a test of the same image shot from f 22 to f 64. The only difference I can see is in depth of field. The sharp areas are very sharp in each one. If I read the above correctly, the f22 ought to be really superb as far as detail (at least where it is sharp) and the f64 ought to be about 1/4 the resolution - and 1/4 the file the other is. They are both very good.



As Bruce correctly mentioned .... at the POF (Point of Focus), f22 should yield a sharper image than f64. However, if the subject at the POF does not have enough detail, then it will "appear" equally sharp. The difference is, at f22, if you put finer detailed subject at the POF, then you will see how the MTF of the lens and film will yield a sharper image at f22. This where the nitty gritty comes into play. Just imagine a single metal column at the point of exact focus.... no detail....shoot at f22 and f64, it would be hard to tell the difference in the processed film....at best you will see tiny differences at the edges under high magnification. Now put a Pine Tree at the POF, you will see the difference very handily.... Of course, this is just at the POF....at the near / far points, as you suggest will vary greatly based on f stop.



> This is where I have trouble - the numbers don't match my experience.


This is because you are not applying all the variables to the equation. This is the value of performing tests with targets, it removes the subjectivity of saying, "they all look good"... with targets, you see where the cut-off is at each f stop and camera to subject distance. When done right, I can assure you after years of testing, the reality of recording detail is very predictable. But I will agree, it can be tricky for the newb... lots of variables within the optical principles and lots of variables in the particulars of specific lenses, films, etc. Again, the goal is not to nail results to within 5%....the goal is to prevent us from being off by 500%.... and thinking we are right... that's what the math is for... it steers you in the right direction for a specific task.



> I'm not trying to be argumentative. I just don't see it.


I understand you are not being argumentative...no problem. I hope my explanations helps you understand why "you don't see it" .... shoot test targets, and you will discover the basic math is remarkably accurate.



> This leads me to the conclusion that, as with many numbers, they are theoretical. Imacon is famous for claiming that they have a DMAx of 4.8, which is beyond the maximum density of film.


When doing a test, you need subjective information and must understand all the variables. From what I am reading, that does not seem to be the case here... the math is not "theoretical" it's been proven science for 50+ years. It's easy to be fooled by a test, then assume the math is faulty. I would not compare sound science like this (with no monetary motive behind the science) with Imacons marketing claims of x dpi or y Dmax. We know manufacturers exaggerate, stretch the truth, deceive, etc. to help sell products. This is a weak analogy.



> However, i have a 8x10 image, scanned at 2000 ppi I am looking at that is just over 300 mgpxels (16000 x 20000), about 500 megs of B&W and I don't see how it relates to being only 1/3 usable pixels.


The mistake you make is.... you assume the dpi you scan at - dictates the amount of resolution a given piece of film contains. You have the tail wagging the dog. The amount of resolution film contains is determined during imaging, not during scanning. I have seen you mention this several times. Here is a simple example that hopefully will bring these numbers into reality.....

We shoot a flat target at f45 with 8x10 color film with 60 lp/mm MTF at the given contrast value of the target. We assume perfect film alignment, focus and diffraction limited lens at f45, single focal plane.... OK?

AT best, we can resolve:

1/(1/33+1/60) = 21 lp/mm

(1500/45 = 33 lp/mm aerial rez)

This means a 30 lp/mm target at the film plane will NOT be resolvable under magnification...it will appear as a blob, vs. cleanly defined lines. It means 20 lp/mm target on the film will appear barely discernible....not a blob, but not sharp. A 10 lp/mm target will look sharp, and a 5 lp/mm target will look razor sharp. This is MTF at work... the finer the detail, the less contrast that is Transfered (hence the T, in MTF) This testing procedure is as old as the hills.... no voodoo math here.

Now, how much "usable resolution" does that 8x10 film have? The film has nothing resolvable over 21 lp/mm.... so it's resolution is stated as 21 lp/mm. Keep in mind, it might have areas of much less resolution, but 21 lp/mm represents the max. recorded resolution.

But since we want MP's, we must consider the film area.

21 * 2 = 42 lines per mm (not pairs)
lines in X Y axis are analogous to pixel

8x10 film is 200 x 250mm

or 10,714 * 8,571 = 91 MP


So 91 MP is the total amount of resolvable information in that 8x10 piece of processed film. No scanning is required to determine this... only a good film loupe, microscope and some simple math. It does not matter how deep you scan that film, there can not be more useful information than 91 MP. The scanner, AT BEST, will only extract the useful information in the film, it will NOT create new information. Hence why there is NO relationship between the size of your scanned files and the useful MP's in the film. Larger scanned files are due to the inefficiency of the scanner, which includes all the variables I mentioned in my previous post. Make sense now?

End of Part I

[bglick]

Part II


As a side note.... this 91 MP would be reduced to 57 MP, if we used f64 instead of f45...sad, huh.... Also, the 91 MP value would be be lessened by shooting a scene with depth, as the near and far points will not resolve as high as the POF. But that's another can of worms. Hopefully now, you can see how these recordable resolution values are not in the 586 MP range, determined by reverse calculating the MB's of a scanned file. Again, the best analogy for that practice is the "tail wagging the dog." You approached the equation from the wrong end.


For a quickie digital comparison, the f stop used, the depth and the pixel count are all variables that must be considered. But in keeping with normal f stop ranges for each format, as a general rule, using these smaller format sizes, which enable shorter fl lenses, hence lower f stops, a digital capture will produce 50 - 75% of recordable MP's, vs. their MP value. So a 60MP back can produce resolvable MP's in the 30 - 45 MP range. But these good digital backs have the ability to uprez very well, which will add some useful MP's back. The range accounts for color of targets, f stop used, lens type used, grid pattern of subject (sag. vs. tang.) etc. etc.


As you can see from this, its very feasible the new 60MP backs will match 8x10 color film (or come damn close) when 8x10 requires f64 to accommodate the depth in a scene. But 8x10 in B&W and / or infinity focus will blow away the 60MP backs.


Before digital, there was no reason to level the playing field between the two recording medias, so we only referred to resolution in lp/mm. But since MP's are all the rage today, everyone wants apples to apples comparisons between the two, so we are forced to convert OFR (On Film Resolution) lp/mm - into MP's, vs., MP's into lp/mm. IMO, the real benefit of this method is..... using MP's of recorded resolution accounts for the "area" issue between all the varying size sensors....it also levels the sensor density, film MTF's, lens MTF's, f stops, etc.

[bglick]

> But don't forget, the image produced by a lens in front of a digital is also degraded by diffraction.


Agreed Ron, we cross posted..... you will see this is accounted for in my digital values...much easier to equate MP of digital capture as 50 - 75% of the native pixel count, which takes many factors into account, not just 1/R.


Also, for your MP values, they represent the true MP values for flat subjects or infinity focus.... for depth, these MP values depreciate significantly...

[rvhalejr]

...
rvhalejr> There cannot be any meaningful objective digital comparisons unless S/N is taken into account...
bglick>>I don't think S/N ratio is an accurate description...
Rich, you are mis-applying my comment.

Sorry, just an old bit-kicker trying to drill down and attenuate the noise floor .,.

S/N ratio is a component of scanning film. However, the reason the scanned file is so much larger than its digital capture cousin, is due to many other MORE significant factors. Such as optics, lighting, mechanical movement, object sizes on film, as well as the Nyquist principle of over-sampling.

Yes, and scanning backs have similar internals and fair comparisons could be made with those (to both work flows). Perhaps he biggest difference between film and the digital back work flow is the initial high IQ (and high S/N ratio) as a result of its direct measurement and the proprietary software/firmware being run inside the digital back (on the SOCs, ASICS, FPGAs, etc.).

It may have been mentioned as of late that they may have started using optical cut-off filters for frequencies above the devices photo-site spatial limit (sub Nyquist diffusion filters). This might be considered the photo-site array equivalent of using fluid in the film scanning process (brick wall noise reduction without losing very little, if any, image detail i.e. signal). It would seem capable of attenuating noise, spurious resolution, aliasing etc.

Why not let the digital back processing cycles be equivalent to the fluid scan post-processing software steps on the desktop, they both reduce noise, both interpolate, up sample, sharpen and so-on

If you re-read what i wrote, I suggested that converting a scanned file MB's into MP's is voodoo math....there is NO sensible relationship between the MB's of a scanned file and MP's of a digital captured file.

Ok, I'm no on-the-metal chip expert, but some bit kickers/firmware/software engineers might have a slightly different perspective (as I will try to explain soon).

Many people mistakenly promote this fallacy.

Ouch! MB's of a film scanned file have a MOUNTIAN of noise, fluid can be used to attenuate grain (noise) quite nicely. Then MBs need processing through a suite of desktop software to get anywhere close to the digital IQ output (and hopefully stay on a level playing field).

Instead of creating a NASA level project from these comparisons,

Comparisons between the work-flows can be fast and simple (arguably over simplified here, ignoring gamma, standard deviation of the noise distribution, etc.).

To illustrate one might look at the image levels distribution curve in PS after selecting a 100x100 pixel square of blue sky (store a screen snap on the side).

Pick another square 100x100 square next to it, strong Gaussian blur the second square, look at the image levels distribution curve and you should have nothing but image data (ideally perfect with very little noise or grain).

Compare the two, S/N all done

its easier to view through a microscope what your film can resolve...that's the true resolution, and represents the MP's of IQ the film contains. (very general statement, as resolution can vary across film)....

GAG !!! Away bad Microscope, Away !!! (Otherwise it all looks like highway blacktop with your nose pressed against it). There is very little film IQ until (in PS on the desktop) noise is eliminated at the level just below the least detectable piece of an image under high zoom, then (as a minimum) down sampled and slightly sharpened.

If done right it becomes very difficult to determine what work-flow the image came from, but I'm old, so maybe not.,.

... the IQ MP's / MP's converted from scanned MB's , represents the scanner efficiency. It will range from 25% - 90%.
I can't fathom why anyone in the photography world would need to dissect each component of the over-all scanner efficiency value..?

Amen.

Got it?

A little DSP here, a little DSP there and pretty soon it all may have a fairly similar IQ (I've neglected white balance, color drift, etc.) at least for old people (without a 35x loupe) .,.

Perhaps, if one allows for a level playing field (heavy post fluid film scan processing in exchange for the backs equally powerful interpolation, noise reduction, sharpening firmware, etc.) a useful educated comparison might be made.,.

[bglick]

> Why not let the digital back processing cycles be equivalent to the fluid scan post-processing software steps on the desktop, they both reduce noise, both interpolate, up sample, sharpen and so-on


These comparisons do not consider post processing...if they did, digital would gain a further edge. It's the nature of processing, cleaner pixels up rez better.... when you toss this mix into the comparison, you can often add an additional 10 - 15% of the recorded MP of the digital capture...... just an estimate of up rez digital has.

In most of your posts you mention fluid scans. Most people on this forum doing high rez scans all fluid mount. So it's always assumed, vs. a new insight to squeeze more resolution from film. But regardless, the comparison always assumes the scanner retrieved 100% of the recorded resolution from the film.



> There is very little film IQ until (in PS on the desktop) noise is eliminated at the level just below the least detectable piece of an image under high zoom, then (as a minimum) down sampled and slightly sharpened.


This is true of both digital and film.... sharpening digital files is an integral part of the work flow, it's not really an option. But I would not suggest there is very little IQ in film, as 10 - 15x darkroom enlargements were the norm for years... of course, no digital manipulation then... that is a LOT of IQ ....



> Perhaps, if one allows for a level playing field (heavy post fluid film scan processing in exchange for the backs equally powerful interpolation, noise reduction, sharpening firmware, etc.) a useful educated comparison might be made.,.


This "IS" the comparison being made, today and the past 10 years now....... Once again Rich, as I mentioned many times in your other thread, fluid mounting, digital processing of the scanned film is all considered "normal" work-flows....its not the new breakthrough that you continue to suggest. So what you refer to as an "educated comparison" is exactly what has been discussed in this thread. As mentioned, if you kick up digital a bit for its edge in up rez, it simply closes the above gap a bit more... not enough to rock the comparison. I am trying to give film a fighting chance :-)

[neil poulsen]

. . . Thank you for sharing your discussion with Schneider technicians. I wasn’t aware of that statement.

My question here would be: why not spare a reasonable amount of money and buy a 1Ds III or D3 instead of that 22Mpx back you are speaking about?
I read you plan to use such a back on your current 4x5” camera or on a 6x9 arca.
But is it the only reason?

I hope there is a quality step between 22mpx 35mm digital cameras and 22mpx digital back. And I understand it is your bet here.
Moreover I imagine that you plan to use movements on such a digital back in order to optimize the sensor native file size instead of stretching 35mm cameras file with perspective tool on Photoshop (like A LOT of photographers do).

But what is the reality of quality question here?
How leaf or hasselblad hope to sell more 22mpx digital back at such prices when 20mpx 35mm cameras are currently buyable second-hand?

I am aware that I’m exaggerating a bit, but you got the idea.
I am curious to hear your point on that. . . .
CA.

Hi. This has been an excellent thread.

There are three reasons:

>> I want the movements that a digital back on a view camera can offer.

>> I regard a 35mm DSLR chamber as small, and light bounces around in there. The flare from an image with overcast sky, or in snow, etc., is PAINFULLY obvious. It's really awful. A digital back on a view camera can deal with this.

>> I've concluded from my Kodak SLR/c that has a full-frame sensor that the color integrity is better than on other DSLR cameras. My camera does not have an anti-aliasing filter, which is the same case with digital backs. Almost all other DSLR cameras have anti-aliasing filters. The only other exception of which I'm aware is the Leica M8. I find this very interesting, given Leica's preoccupation with producing a quality image. On this point, I've not myself done a side-by-side comparision. I draw my conclusions from people who have done these comparisons.

I might add that a second advantage of not having an anti-aliasing filter is improved resolution. Of course, not having the aa filter leaves one open to moire, which can be a real problem for some images. But, not so much for landscape.

------------------------

To add a little context to my comments, I do all all my photographing on a tripod. I rarely photograph handheld, DSLR or otherwise. So, I don't need the speed that a DSLR can offer.

[Stefan Lungu]

Strange how people see things differently. I also have a 1DsmkIII, I'm sort of disappointed with it, not that it does not perform very well, it does mostly. It leaves me thinking it should be better, hence a return to LF for me for the fun side of photography. The more pixels I have in 35mm cameras the more I like film. I've not found any Canon lenses to get me excited either. The latest range of pixel packed 35mm just leave me cold, all those pixels help highlight digital faults and make me realise how good film is and how I took it for granted.

Kevin.

Now, I don't think getting the 1Ds MKIII is not a good idea, unless you have the lenses to serve that fully packed sensor. So a dedicated photographer like yourself will be disappointed with most of Canons lens offering. But the talk here made me think : I started using a Canon 300D, 6Mp sensor. Now, to be able to double the frame length, I need four times more pixels, and that would be around the new 1Ds MkIII or 5d MkII. That is a hell of a leap in years and price, not to mention the lenses that looked "ok" or "good" on that sensor compared to how they would look like on the 20+MP sensors. I think a lot of the problems come from this never ending pixel race that brings nothing but trouble to the end user : artefacts from the noise reduction that has to be bigger with the smaller pixels, larger files and better lenses that are needed to get something right on that sensor.

[bglick]

> I think a lot of the problems come from this never ending pixel race that brings nothing but trouble to the end user : artefacts from the noise reduction that has to be bigger with the smaller pixels, larger files and better lenses that are needed to get something right on that sensor.


This is soooo true, well said..... its like Pandoras box. :-) However, part of the discussion here addresses this.... don't let the pixels get too small, by increasing the size of the sensors...you still need good lenses, a step better than most LF lenses, but its a very good solution... the deterrent is only price, such as the $40k P65, 60MP backs. The big question is, will these ever become reasonably priced?

HD Tv's were $15k in 1997, today, $1k.... We have all become accustomed to these technology price drops...even Sony at $3k for 24MP is setting new price bars. Whether this happens with larger, lower volume sensors...who knows... my guess is, the volume is just too low to ever get the price reductions that will make MF backs more mainstream.

[rvhalejr]

> as I mentioned many times in your other thread, fluid mounting, digital processing of the scanned film is all considered "normal" work-flows....its not the new breakthrough that you continue to suggest.

I cannot recall ever thinking that a fluid scan might be considered a breakthrough.

That word was once used in reference to a possible breakthrough coming soon[1].

When fluid scanning has been spelled out it has been for clarification, not intended to be overly redundant. I'm surprised that anyone would be annoyed by its use.

I would find it very interesting if you were to kick off a public discussion at some point about what you seem to refer to as "scanning at depth" in film.

I could be wrong but that sounds like your referring to the scanning the discrete color layers within the emulsion. Yes, that is an over simplified statement by design.

> I am trying to give film a fighting chance :-)

And so am I .,.

[1.] Hence research into one facet of fluid scanning that very well may prove fruitless, but those details might be better discussed off line .,.

[bglick]

> I would find it very interesting if you were to kick off a public discussion at some point about what you seem to refer to as "scanning at depth" in film.


I simply meant, scanning at greater ppi.... this is often referred to as a "deeper" scan, meaning, trying to extract more resolution. The Depth of Field of scanning lenses will not allow focus on each level of color within the film.