Top-end digital concerns

[sanking]

"andy you will have to do your own research on this.... really, not only have I spent hours posting all this, now I am supposed to be the R&D support system for everyone who wants supporting information for any information they are not sure of ?? I guess thats the thanks ya get on internet forums... sheeeeesh"


I have done my own research, and a lot of testing scanning resoution targets as well. My experience is absolutely not supported by your claim that it necessary to over sample at 10 lines per mm to get 5 lines per mmm on the print. It may be that your claim is supported by some previous testing, which is why I asked if you could point me to a study. What do your studies, assuming you must have made some, show?

You are the one who is claiming the devil is in the details. I simply am asking for some details to support a claim that you made.

Sandy King

[sanking]

[QUOTE=bglick;431718
Place a 100 lp/mm target on your high end scanner and scan at that same resolution, see if you can resolve it....if it fails to resolve, see how much higher you must scan to resolve it... that will advise you of your scanner eff. at that spatial frequency. You will find at low spatial frequencies, scanners are very eff....at higher frequencies, they become less eff....., just like the MTF of lenses and film in the photographic world....

[/QUOTE]

Not many people have a high end scanner that can resolve 100 line pairs per mm so that test is off limits for me.

What I prefer to do is look at the resolution target on film with a microscope. Generally 30X-40X is high enough magnification, but you might need more for some fine grain B&W films. You can immediately tell from this exercise how much resolution the film is resolving with the camera system and development used.

Now scan the target and see how how you will have to set the resolution to caprture what you can see with the microscope. In work I generally over sample a bit, but I don´t find it necessary to oversample by 2X, not even close.

Sandy King

[bglick]

> I have done my own research, and a lot of testing scanning resoution targets as well. My experience is absolutely not supported by your claim that it necessary to over sample at 10 lines per mm to get 5 lines per mmm on the print.


Sandy, please re-read what I wrote. I did not suggest that all scanners need to over-sample by a factor of 2x to extract film data. I made it CRYSTAL clear, the better the scanner, the more efficient. The lower the spatial frequency of the target, the greater the efficiency of the scanner.


In your case, you have a very high end scanner, I did not suggest it would require 2x oversampling.... so I don't dispute your findings at all, they agree with my findings with all my scanners through the years.... As I made it clear, I used 2x as an example in the "printing" post, for simplicity sake. Then I suggested everyone test their own scanner to determine the efficiency....your testing is with your HIGH-END scanner.....but what about a $99 film scanner, will it perform as well as yours? Get it? I was simply laying out the protocol so others could get a feel how there scanners measure up....that post was not addressed to you specifically....IT WAS GENERIC!


In my case.... with low spatial frequencies, I can scan within 10% of the film rez. AT very high frequency on the film, say 60 lp/mm, I scan at about 1.3 - 1.4x the film rez. As I mentioned previously, every scanner will vary based on the hidden firmware that dictates the scan process and the specific characteristics of scanners optical chain.



> You are the one who is claiming the devil is in the details.


yes, the devil IS in the details, and I am suggesting to determine your specific "details", TEST YOUR OWN SCANNER ! I never suggested that I possess ALL THE DETAILS OF EVERY SCANNER EVER BUILT ON THE PLANET ?? You gotta be fair, so this clearly deserve a double sheeeesh...

sheeeesh.....sheeesh....



> Now scan the target and see how how you will have to set the resolution to capture what you can see with the microscope.


Sandy, do you realize, your testing procedure above is repeating the protocol I wrote from the beginning? So whats the beef? It's that simple, - test IT! If you go back in this thread, I suggested a range of over sampling that may vary from 1.2x - 3x based on the quality of the scanner and the frequency of the target... I am sure a $99 scanner even at 3x over sampling will not extract high frequency targets.... hence the HUGE range I tossed out as an example, due to the WIDE range scanner quality at all many different price points...

Are we on the same page yet?

I feel like I am in a Twilight Zone episode...

[Gordon Moat]

bgclick, or whatever your real name is, I refer back to the origination, and some individuals comments:

http://www.largeformatphotography.in...12&postcount=3

http://www.largeformatphotography.in...72&postcount=5

http://www.largeformatphotography.in...5&postcount=26

I am left curious as to your background, though I suspect engineer; which might explain the fascination with numbers. Your quoting method is so different from that of anyone else who posts on LF Forum, that it has become tedious to respond to you.

I suggest you research people like Sandy King prior to suggesting their methods are improper, or that they lack thoroughness. While I rarely agree with Lenny Eiger, I am finding similar results to his in real world images, namely that your numbers just don't match the image results out there; basically what you introduced into this thread about the issue of Image Quality.

http://en.wikipedia.org/wiki/Digital...lm_photography

I suggest you tackle this Wikipedia article, because your efforts will be better spent addressing the citations needed areas. Likely you will find a larger audience.

http://www.galerie-photo.com/film-co...esolution.html

I don't know if you read French, but I greatly respect the views of Emmanuel Bigler. Galerie Photo is also a very informative and extensive website, which I highly recommend.

Finally, if I were to accept your figures at face value, they are at odds with the reality of actually seeing highly detailed prints from Gursky, Burtynsky, et al. Perhaps these photographers have beaten the math.

Ciao!

Gordon Moat Photography

[bglick]

> I am left curious as to your background, though I suspect engineer;

Guilty as charged... but I have been involved in photography and optics for over 30 years. Not just camera optics...



> Your quoting method is so different from that of anyone else who posts on LF Forum, that it has become tedious to respond to you.


This explains why we are struggling so much with the MTF stuff :-) if you see ">", that's a quote from you.... everything else is fresh, from me... OK? No one else seems to struggle with this...




> I suggest you research people like Sandy King prior to suggesting their methods are improper, or that they lack thoroughness.


I am back in the Twilight Zone episode.... are you reading what I WRITE? I did not suggest ANYTHING was improper with Sandy's protocol, his suggestion repeated EXACTLY what I had written previously, hence the confusion I had. I get the feeling no one is reading what is written in these posts...which is a huge source of frustration....


the Wikipedia link is a very basic primer on digital vs. film... not much to disagree with... other than, in our examples here, we are leveling the playing field between all the formats, by stating "recorded resolution in MP's" vs. native MP's. We have done this for both film and digital....this accounts for 1/R AND the recording media size...the two factors that dictate total resolution. It's my suspicion, you have gazed over this part.... this effects the numbers greatly... I could be wrong, just trying to find the parts you disagree with as NEVER mention specific faults...

I don't read French...sorry...


> Finally, if I were to accept your figures at face value, they are at odds with the reality of actually seeing highly detailed prints from Gursky, Burtynsky, et al. Perhaps these photographers have beaten the math.


If Gursky and Burtynsky have overcome the laws physics, the laws of imaging, they should be nominated for a Nobel Prize. Really.... no joke there. I can assure you, if they have defied the basic laws of imaging, Zeiss, Leica, Schneider would offer them million$ for their discoveries. :-)


Gordon, as I am sure you would agree, our communications is failing, and my fingers are tired of typing. I explained everything in great detail.... I have given you ample opportunity to suggest where the faults are....and yet, you have not offered one bit of information that is conflict with basic imaging laws. IMO, you have added fuel to the fire of, THATS NOT WHAT I SEE, and FILM RECORDS 400 MP, all with no details...etc. The problem, as I pointed out many times, is without details of "what you see", your position can not be evaluated. This is the value of "seeing" and testing with targets.


I am sure you see great prints, but there is a lot that went into that great print, and without all the details, its meaningless to introduce such into these discussions. The testing procedures are simple, you are welcome to try them any time, it costs just a few dollars.... so why not put the rubber to the road and get to the bottom of this? Again, your conclusion is, yourself and Lenny have proven Kodak, Fuji, Leica, Zeiss wrong, by a factor ~10x.... (under normal or avg. photog scenarios) of course, no justification, no tests, it's all based on what you see in certain images... too many variables to make constructive comparisons.


To close this issue out, as this is getting way too repetitive.... I would leave you with two statements that demonstrate how the range of details (which are usually left out of the discussion) can span the gamut, which is why this issue is all over the map..... These are two extremes Statements, which both "can" be true.... I won't repeat the details of the math, as its all in this thread...


Film vs. Digital capture


STATEMENT #1 - "FILM CAN RESOLVE 74x GREATER THAN DIGITAL"



(I purposely leave out details, cause therein lies the problem)

Film capture details...
8x10 B&W film 100 lp/mm MTF, diffraction limited f11 lens, highly detailed subject, infinity focus, tight camera, excellent processing, etc.
This scenario we will record 666 MP of resolution. (this is feasible)


Digital capture details
24 MP DSLR with slightly aberated lenses delivering f11 aerial rez...
This scenario will record 9 MP of resolution.

666 / 9 = 74x yep, it checks out, this Statement can easily be justified... the luster of LF film is still alive n well !



STATEMENT #2 - MF DIGITAL CAN RECORD 7X MORE RESOLUTION THAN MF FILM.

Film Capture details
MF 6x4.5, ISO 400 color neg film 40 lp/mm MTF, shot at f16 for DOF...
This scenario will record 8 MP of resolution

Digital Capture Details
P65+ back, Digitar lenses at f4 (diffraction limited), shot at infinity...
This scenario will record 42 MP of resolution.

42 / 8 = ~ 7x Yep, this checks out too, statement can be justified.



Get the drift here? This is why this debate is so insane... the answers run the gamut, from the first Statement where film BLOWS away digital by 74x .... and the second statement where MF digital blows away MF film. How can both be true? It all comes down to the DETAILS of each captures variables......, f stop used, DOF, subject detail, film type, format size, focus distance, ... etc. etc. etc. If I introduced view distance in the equation, I could really swing the numbers even further apart. AS you can see, the details can swing the outcome in either direction. And this happens in the real world, you get people comparing images without the details, and they come to incorrect conclusions.


This is why, when I hear statements like, "thats NOT what I see" .... I have to ask... what distance were you viewing from, what printing method was used, what capture method, what film or sensor, what lenses, what f stops, how much DOF, etc... only then can we make sense from "what I see" comments. Then of course, you need a sensible comparison to do a comparative analysis. For the newb, yep, this can be tricky stuff, lots of variables. But at least one photog on this forum digested this entire thread in just a few hours.... and has his Excel spreadsheet cocked and loaded :-)


Gordon, you call this "being obsessed with math." I respectfully disagree. In my world, I would classify this as "implementing 6th grade math to pre determine the outcome of my imagining." It's a simple and powerful tool to plan a shoot. I can determine what format too shoot, what lenses, how to scan, how to print, what paper, what view distance, etc. etc. It takes me 10 minutes to run a scenario, thanks to spreadsheets...its no big deal... It's basic imaging fundamentals reduced to a handful of variables.... quite elementary stuff......and the best part is... its proven!


Now, you should be happy, I justified your position, film records 74x more than digital! :-) yeee ha!

[dh003i]

Film capture details...
8x10 B&W film 100 lp/mm MTF, diffraction limited f11 lens, highly detailed subject, infinity focus, tight camera, excellent processing, etc. This scenario we will record 666 MP of resolution. (this is feasible)

First, do you mean 100 line pairs per mm, or 100 lines per mm? I know that your nomenclature was for line pairs per mm, but 100 lp/mm seems really high.

And even 100 l/mm seems high for a lens covering the entirety of an 8x10 lens at f/11. The Nikkor-SW 90/8 (and presumably 90/4.5 as well) has a measured resolution to BW film of 80 across the board at f/11. That's (80 x 25.4)^2 x 8 x 10 / 10^6 = 330 megapixels. But I'm not sure if Chris Perez measured it far enough out where it would need to go to cover 8x10 (it does cover 8x10). Does "edge" resolution mean edge of the film without any shift, or the lens shifted such that you're measuring the resolution from the very edge?

[dh003i]

Regarding the over-sampling a scanner needs to capture all of the resolution of film, doesn't the 1/R formula imply that a scanner can never capture all of the resolution of film?

I.e., a 6000 dpi scanner corresponds to 6000/26.4 = 236 l/mm. Assuming an image recorded to film at 80 l/mm, the final resolution on your computer would be:

R = 1 / (1/236.22 + 1/80) = 59.76 l/mm

and that's over-sampling by almost 3 times. Granted, it is very very close...but without over-sampling (as some people would think you can do), you would get

R = 1 / (1/80 + 1/80) = 1 / (1/40) = 40!

So without any over-sampling, your final resolution is cut in half!

Take a look at these numbers (calculated the same way as above); it really sucks:

Over-sampling ; Actual Resolution (of scan)
1.0 x ; 20.6 Mp
1.1 x ; 22.7 Mp
1.2 x ; 24.6 Mp
1.4 x ; 28.1 Mp
1.8 x ; 34.1 Mp
2.6 x ; 43.1 Mp
4.2 x ; 53.9 Mp
7.4 x ; 64.1 Mp
13.8 x ; 71.8 Mp
26.6 x ; 76.7 Mp
52.2 x ; 79.5 Mp

But the resolution of the film itself is (80 x 25.4)^2 x 4 x 5 / 10^6 = 83 Mp.

[dwhistance]

I have followed this thread with interest as I have the previous similar threads, albeit that it is clear that the participants are tiring of it from recent posts.

I agree with, and understand, the math used by bglick and wholeheartedly agree that "THE DEVIL IS IN THE DETAILS", however I think this is exactly where many of these discussions go wrong. This is because there is a tendency, at least in the mind of the reader, to compare say the MF digital back used with a lens at optimal aperature with 4x5 film with the lens well stopped down. ie apples with oranges. This is similar to the medium format example given above although I realise that bglick used it to make just this point.

I also feel that not enough thought has been given to the nature of the pixels in the digital back, particularly in one shot mode, where even the P65 doesn't really have the stated number of "real" pixels but a greatly reduced number due to the Bayer pattern. Quite how many I am at a loss to say but the reduction must be significant given that you essentially need a block of 4 pixels to generate accurate colour. Simplistically that would suggest a reduction of perhaps 50% to 75% although it would be good to know what it really is. I know this will raise all sorts of comments on the wonderful algorithms used in the processing, however that doesn't change the fact that interpolated data, however well it is done, is not the same as real data - we are essentially back to the point bglick made, correctly, about scanning.

Perhaps the impact of the Bayer pattern will go some way to explain the empirical findings of Gordon Moat, which I agree with, that in real life (landscape) prints 4x5 film appears to better MF digital backs by a wider margin than the calculations would seem to suggest. It would also explain why, when used in the studio in multi-shot mode these same backs can hold their own against larger film and even some scanning backs.

David Whistance

[sanking]

I agree that the quality of a pixel as it relates to film resolution is something of a black hole. The only way I know to compare the two is with specific testing of resolution targets, taking into consideration the angle of view of the lens and the size of the film versus the size of the sensor. You can get the results of resolution with digital capture directly, just by looking at the image on your monitor at high magnification and doing the number. Film is much more complicated because it is dependent on both the film (color or B&w, grain, ASA) and the quality of the scanner.

As others have said, there is no one magical answer to this. It really depends on the details.

Sandy King

I

I also feel that not enough thought has been given to the nature of the pixels in the digital back, particularly in one shot mode, where even the P65 doesn't really have the stated number of "real" pixels but a greatly reduced number due to the Bayer pattern. Quite how many I am at a loss to say but the reduction must be significant given that you essentially need a block of 4 pixels to generate accurate colour. Simplistically that would suggest a reduction of perhaps 50% to 75% although it would be good to know what it really is. I know this will raise all sorts of comments on the wonderful algorithms used in the processing, however that doesn't change the fact that interpolated data, however well it is done, is not the same as real data - we are essentially back to the point bglick made, correctly, about scanning.

David Whistance

[bglick]

.

> First, do you mean 100 line pairs per mm, or 100 lines per mm? I know that your nomenclature was for line pairs per mm, but 100 lp/mm seems really high.


lp/mm = Line Pairs Per mm. You are confusing the MTF of the film, vs. the recorded OFR. (On Film Resolution) Remember?



> And even 100 l/mm seems high for a lens covering the entirety of an 8x10 lens at f/11. The Nikkor-SW 90/8 (and presumably 90/4.5 as well) has a measured resolution to BW film of 80 across the board at f/11.


The same 1/R math applies ....

1/(1/(1500/11) + 1/100) = 58 lp/mm recorded to film
(not 100 lp/mm)

58 x 2 = 155 lines per mm

200 * 250mm for 8x10 film = 666 MP


> Does "edge" resolution mean edge of the film without any shift, or the lens shifted such that you're measuring the resolution from the very edge?


Edge rez, equals the resolution recorded at the edge of the film, in this case, usually the 8" edge, or 4" radius from center. I only used this f11 lens as BEST CASE scenario for the 810 film.... I am sure there is a dandy f11 lens optimized for infinity focus, as you have already found some vintage lenses that perform near diffraction limits for a specific focus distance and a given f stop....


Your scanning comments are valid, however, there is tooo many variables to reduce this to math, as you are forced to make too many assumptions. Hence my suggestion for simple tests to determine the answers you are pursuing... a few generic comments:


1) You never know how much over scanning is being performed by the scanner, as it's common for scanners to over-sample and reduce file size without your knowledge.

2) Using 1/R for scanners, you must treat it the same as our camera imaging system, you must enter BOTH the MTF of the lens (just like a camera imaging system) and the MTF of the targets. Just as a side note....why do expensive scanners perform so well? A diffraction limited f4 - f5.6 macro lens such as a digitar, cost $2k+ just for the lens, not to mention all the mechanicals, lighting, electronics, software, etc. So for lower end scanners, with lower end lenses, often times high frequency targets can NEVER be resolved, regardless of the dpi advertised on the box....its just native pixel count, which is only one component of 1/R. This is why the entire issue, unlike camera imaging, is a pandoras box.... there is too many variables we do NOT know, hence the simple scanner test both I and Sandy proposed, will determine the capability and efficiency of a specific scanner.... in effort to prevent this thread from turning into a scanner thread, I will stop there...


> Quite how many I am at a loss to say but the reduction must be significant given that you essentially need a block of 4 pixels to generate accurate colour. Simplistically that would suggest a reduction of perhaps 50% to 75% although it would be good to know what it really is.


1/R math is basic imaging math.... it is pre dates digital capture. However, digital sensors have their own MTF, although we are not provided specific MTF data, the pixel density and sensor sizes are clearly defined. Your point regarding Bayer RGGB pixel layout is valid. Bayer sensors introduce this added variable of color pixel sites.


Without getting too involved into digital recording..... here is the reality of the RGGB pixel layout as it relates to 1/R.....this is from my own experience shooting color targets. There is 3 general classifications....


1) When shooting black targets, 1/R prevails, remarkably accurate. Because all pixel sites will "see" the B&W targets.

2) When shooting Green targets, 1/R outcome is "nearly" identical to B&W targets.

3) When shooting R, B targets, there is clearly a fall off, in the 25% range.


In addition to color variable, the fixed grid pattern of digital sensors, does not record mis aligned targets as efficiently as those targets on the same XY grid orientation. At high enough frequencies (dependent on sensor density) 45 deg angled targets create new degradation issues with digital capture (vs film). Target orientations that align with the XY sensor pattern perform as expected, no losses. However, targets at 45 degrees to the sensor grid, DOES introduces added degradation, BUT again, only above certain frequencies. So, based on the subjects you shoot, this "can" be a very unwelcomed shortcoming.


And yes, flat bed scanners are vulnerable to this same orientation issue, hence the need to over sample. Drum scanners are almost immune to this issue, due to their non XY recording.


When I first start testing digital cameras, I expected huge losses due to these potential shortcomings....but in reality, they are very minor.... the reasoning is two fold.... first, many colors are a mix of several colors, which allows the pixels to acknowledge the existence of colors which are not a perfect match to the pixel recording color. Secondly, the brilliance of the interpolation algorithms in the camera are nothing short of amazing. Other than pure RED, or pure BLUE targets (specially at 45 deg. orientation) the Bayer sensor technology performs almost as well as a Foveon sensor where each pixel site records RGB. Its VERY easy to test this with color targets, placed at different orientations throughout the recording area. I made a long post about this in more detail, in another thread awhile ago - if anyone is interested....but that's the short story.



> I know this will raise all sorts of comments on the wonderful algorithms used in the processing, however that doesn't change the fact that interpolated data, however well it is done, is not the same as real data


If you ever tested this, you would be shocked at how effective the entire working system is. Because all these pixels "see" small components of other colors, the software really does an excellent job interpolating....BUT, its not a 100% guess as some suspect, as it does have "some" data to work with.....so the "PURE" interpolation is limited to distinct color subjects, mainly R or B. Which if a subject is 100% of one these colors, it would reduce the "effective" native pixel count by 75%. In which case, the resultant recorded resolution does take a serious hit.



> which I agree with, that in real life (landscape) prints 4x5 film appears to better MF digital backs by a wider margin than the calculations would seem to suggest.


I shoot a lot of 4x5 Velvia at f16 with my SS XL 110 and 150... Excellent 4x5 lenses. The DOF fits within these shots, as I pick n choose shots carefully. This will record ~ 70 MP of usable data (MTF of the film at 70 lp/mm) Compare this with P65+ with adjusted f stop for format size, it will record 31MP (at best due to color and orientation issues). With the previous 39MP backs, even less, just guessing, in the 24MP range.


So it's NO mystery to me, why all my 60" prints from 4x5 Velvia exceed MFDB capture, including the newly introduced P65+ (which I have not tested, but the math will surely predict its gains). So IMO, there is no...


" wider margin than the calculations would seem to suggest"


I consider 3x the resolution to be very substantial. IMO, the calculations actually justify the higher resolution 4x5 output.


However, when you begin to stop down on 4x5, (or 810) the gap begins to narrow quickly.... it's a game - success lies in avoiding the fatal effects of apt. diffraction - the larger the format, the faster it reduces recorded resolution.


And you have correctly pointed out, when you begin to stitch a few digital captures together, it's a game-changer... hence why, there is a LOT to consider when making these comparisons. There is also a lot to consider, when selecting your capture weapons for your specific subject matter and output criteria.