My fixation on photon multipliers got the better of me .,.Originally Posted by bglick
My fixation on photon multipliers got the better of me .,.
Lenny, I was reading your marketing material on your web site....
320 MP for scanned 4x5 film... ???
YIKES!
>> For those interested..... 4x5 color film at 60 lp/mm
>> f16 - 58 MP
>> f22 - 44 MP
>> f32 - 30 MP
>> f45 - 20 MP
It was hard to me to believe into the numbers at first.
Then I remembered that I have two pictures of the same landscape to compare
- a 4x5 picture taken with a 300mm NikkorM at F32 on Delta 100 , wet-mount scanned at an optical scanners resolution of 2450 ppi with no sharpening.
- a 1Ds 11MP (with weakest AA filter among all Canon digital line) an EF 135F2L at F8 ISO100 in RAW. This picture covers less of the landscape area so I think it is appropriate for a rough comparison. The goal is to see if a 4x5 @ F32 "is" a 30 MP from a Digital sensor
The 4x5 scan came out to be 12000x10000.
The RAW converted into TIFF with no sharpening ~5000x3000
I up sampled the Canon's file in PS (bicubic sharper) for 4x to make it have 10000 along the longest side and finally to have the elements of landscape of the same size as on the 4x5 scanned image.
I gave some amount of USM to both files to make them both look the best.
Then I compared side by side the similar pieces of them at 100% magnification.
I was shocked to learn that the up-sampled Canon's file (on the screen) subjectively contained the same amount of details as the scanned 4x5. I think partly due to the virtual absence of the grain and noise in Canon's capture. Well I had to add about 5% of noise (a PS filter) to the Canon's file to make it look "closer" to the scanned 4x5. Otherwise I could not stand the grain on 4x5
So the 4x5 >> f32 - 30 MP sounds like the right number.
Last edited by SergeyT; 16-Jan-2009 at 19:14. Reason: typo
> So the 4x5 >> f32 - 30 MP sounds like the right number.
30 MP? Some will argue it should be 320 MP ? Oh what the heck, it's only a difference of 10x. ;-)
Anyway, it's nice to see "some" sanity offered to the digital vs. film issue.
And you are sooo right.... digital's up rez capacity is mind boggling.... of course, the cleaner the pixels, the better the up-rez capacity, hence the value of these larger pixels. If you up-rez a digital capture 2x, its impressive how well it holds up...and at this 2x size, ~ 10% of the total pixels were actually recorded at capture.... 90% of the pixels are interpolated. If someone tried to convince me of this 10 years ago, I would tell them, they were insane.... it's a tough pill for many to swallow, even today.
I find it difficult to believe that the relatively tiny 35mm sensor on a "full-frame" digital SLR can match the final print quality of a 4x5 image, even when used at high f-stops. Didn't you post a graph showing that prints made from 4x5 had higher lp/mm's than thoe from 35mm, even at higher f-stops (although the advantage deteriorated).
See this test of 4x5 vs. the MF P45 back, where 4x5 still has a (slight) advantage over the best new MF back. Granted, this is a relatively flat subject, and I don't see info about the f-stop used for those pics. But it doesn't look to me like the P45 file can be up-rez'ed particularly better than the 4x5 file. It looks like it is maybe smoother with pure color tones (see the color chart, where film's grain shows), but that it has less detail in other areas. Which would, imo, make it a wash versus the 4x5 file for upscaling. (we can't consider the color, as the author told us there was still something funky going on with P45 color calibration).
And that's for a P45 back, not a 35mm digital. I find it difficult to believe that 35mm digitals can produce images with as much stunning detail as this. The latest 4x5 vs. P45 results seem to show the P45 being equal to 4x5, according to the author. (look for the Rematch! P45 vs. 4x5 Velvia section). My impression is that for smoother color-tones and the leaves, the P45 outresolves 4x5 film, but for the connifer needles, the 4x5 out-resolves the P45.
In any event, for ordinary people, the P45 is certainly way way out of their price-range, and so too are Canon or Nikkon full-frame 35mm digital cameras.
PS: There is also some research (above my head), and experimental verification, of designing lenses that over-come the diffraction limit.
> I find it difficult to believe that the relatively tiny 35mm sensor on a "full-frame" digital SLR can match the final print quality of a 4x5 image, even when used at high f-stops.
Clearly, 4x5 scanned film files will out perform 21 - 24 MP digital capture. Of course there is a lot of variables that can push the comparison further in one direction or another. In my previous posts, where I stated the recordable MP's of 4x5...as you can see, its more than 35mm digital.... even at f32 for 4x5, 30 MP... compare this with 21 MP DSLR * 65% (avg recordable rez factor) = 14 MP. At f16 for 4x5, as you can see, the recorded rez doubles to 58 MP.... this is a significant difference. Hence why the comparisons are so "variable dependent". DSLR's have the huge benefit of low f stops, due to their shorter fl's, so apt. diffraction is not so degrading. Not true for 4x5.
The up rez capability of digital capture varies greatly based on the pixels, technique, subject contrast, subject colors, etc. Sometimes the results are amazing, other times, just good...hence why I try to leave this variable out. The previous poster obviously had a great up rez experience in his comparison. I have had similar experience that just floor me.
The link you provided at the LL enforces the sensible MP values of scanned 4x5 film I provided previously. In this example, 4x5 scanned film just barely out resolved the 39 MP back... and this makes perfect sense with the imaging math above....
39 MP * 70% = 27 MP recorded resolution
vs.
4x5 scanned color film ranges from 30 - 58 "recorded" MP in the common f stop range.
So based on this, I would say the Charles Cramers test probably exposed the 4x5 at f22 or f32.... the resultant comparison makes perfect sense... no vooo dooo ...
More importantly, it demonstrates 4x5 scanned film is NOT 320 MP as some posters continue to endorse....and in another thread, I have seen numbers as high as 400 MP.
Also keep in mind, if the subject matter is static, you can stitch a handful of low end DSLR images together and easily surpass 4x5 IQ. This represents the lowest cost, and highest IQ possible....
> There is also some research (above my head), and experimental verification, of designing lenses that over-come the diffraction limit.
Any links? their seems to be no method to overcome diffraction in ONE single apt. But the methods I read about being tested side-step diffraction, by using multi lens set ups, to capture different parts of the subject.... with this technique, you can use much shorter and therefore faster lenses, in the end, you have cheated the diffraction limits through by combining the images via sensor hardware / software. This will be a major boost to imaging....specially astro photography....not sure if it will filter down to still photography..
My . . . look at the time . . . where to start . . . So, one issue is on screen comparison, in that what you see on any computer monitor is a poor representation of printing capability. Such comparisons even reach an extreme of some people in public wondering why an image that looks so great on their HD television cannot make a good giant print. Now if your goal is only to display on monitors or giant televisions, then I fail to see why one would use film at all.
The Ludicrous Landscape comparisons have been discussed heavily in many past posts here, so . . . consider that these are aging baby boomers with eyesight not nearly as good as it was in the past. I think the level of acceptable comparison has reached the limits of eyesight for these individuals.
Simple aspects left untested: put an anti-alias filter onto a film camera, and place a grid pattern over the film. The idea is to turn the capture into discrete blocks of information. Then a later scan would allow these blocks to be more easily interpolated into an upscaled image. Sounds crazy? Yes, so here is the practical way to do that: view the film through a loupe, noting the detail information. Next use a high end digital back or DSLR to photograph the film. Break up the 4x5 piece of film until the capture is close enough to recording details visible through the loupe, sort of like doing a slide copy set-up. Then stitch the results until a large file is made that equals the film capture details as seen through the loupe. Last, count the megapixels.
Then there is the reality check. I have been vastly impressed at the quality of large prints from certain photographers, though the prints of Edward Burtynsky have impressed me the most. These were mostly in the range of 40" by 50", and from 4x5 film, some of them enlargements and others from a machine. Using 304.8 dpi at the 40" by 50" size, would give just over 185 MP, for those who want such numbers. However, I think a better number is the usual visual capability of the average viewer, which is often thought to be near 6 lp/mm, and oddly enough that relates well to 304.8 dpi . . . probably why that (or 300dpi) has been common printing industry specification for a long time already.
Sometimes the implied message is that it cannot be possible to make nice looking large prints from film, because digital capture is so much more advanced. Obviously this is not something anyone has directly stated in this manner, but when we apply such satire to these arguments, then I think it can allow us to be more objective. Any of us who has been to a gallery or museum and seen nice large prints already has one answer for themselves, and all they need to do is figure out whether they want to emulate the techniques and tools, or go their own direction. The prints of Edward Burtynsky impressed me enough that I bought back into large format three years ago.
Ciao!
Gordon Moat Photography
The problem with up-rez'ing something is that you're just making up details. It may look ok, but that doesn't mean it has more detail. And if you say an "up-rez'ed" a digital capture picture by 4x, you also ought to up-rez a large-format film capture by 4x with the same settings, and then compare that. Now, I wouldn't be surprised if in some things, the digital up-rez's better, due to not having grain (but it does have its own noise problems, and imho, the noise on digital sensors [i.e., my 4/3rds] is less appealing than grain noise). The $20k P45 is an exception, which didn't seem to have digital noise artifacts (I've seen demo units for $20k, now that the P65 is out).
I believe the site making the comparison noted that the costs of film will quickly add up for a 4x5 user, making the now-$20k P45 more affordable in the long-run for those who take many pictures. However, I don't know if he considered opportunity cost in that comparison; namely, there's about an $18k difference between comparable systems for P45 and 4x5 (and I'm not even including lenses for the P45). $18k would earn substantial amounts of interest. In short, you have to do a Net Present Value calculation to figure out which is more cost-effective.
The cost of the P45 is fairly obvious: $20k + lenses, all upfront cost. The cost of 4x5 is more complex. It of course includes the cost of the system + lenses (say $2000), then film ($2/sheet), plus the cost of scanning it. But you can't just add that up over the expected use-life of the 4x5, and say "oh, it's more expensive". You have to discount the costs by some opportunity-cost rate of return (say 6%, probably more), as such:
Net Present Cost = C1(1+r1) + C2(1+r2)^2 + ... = C[1 - (1+i)^-t]/i
where C1 = cost in year 1, C2 = cost in year 2, r = i = rate of return or interest rate, t = time, and C = annual cost, assuming the cost is the same every year.
(normally, we'd do "net present value", but it's difficult to quantify the value of having a camera system, although doable; but this isn't particularly relevant here if we assume the image quality of both systems is the same, so the Net Present Benefit will be the same, but Net Present Cost will be different)
Well, 4x5 scanned film may be close to a useful resolution of 320 MP at wider apertures, if the lens is designed for that (i.e., Xenotar 135/3.5, which is sharp wide open and very sharp stopped down just a little). Of course, that is more useful for portraits and abstracts, not landscape (unless it lends itself particularly well to tilt).The link you provided at the LL enforces the sensible MP values of scanned 4x5 film I provided previously. In this example, 4x5 scanned film just barely out resolved the 39 MP back... and this makes perfect sense with the imaging math above....
39 MP * 70% = 27 MP recorded resolution
vs.
4x5 scanned color film ranges from 30 - 58 "recorded" MP in the common f stop range.
So based on this, I would say the Charles Cramers test probably exposed the 4x5 at f22 or f32.... the resultant comparison makes perfect sense... no vooo dooo ...
More importantly, it demonstrates 4x5 scanned film is NOT 320 MP as some posters continue to endorse....and in another thread, I have seen numbers as high as 400 MP.
But I agree, for normal subjects, where you have to stop down even with tilts, due to the various heights of objects you want in focus, 4x5 won't be 320 MP.
Yes, but then that's also an argument against MF digital backs as well, and against full-frame sensors, or even APS-C. Then everyone should use the DSLR I have, the Oly E-3 (or maybe E-520, that's cheaper), with the 4/3rds sensor. With the 2x crop factor over 35mm, you can really peer far into the image with a 300/5.6 lens (equivalent to 600/11.2 in 35mm), and take an enormous capture. Although it would probably be better to use an uber-high-quality lens, like the Olympus Zuiko Digital ED 300/2.5 or Sigma Apo 300-800/5.6 EX DG HSM. Although those cost thousands of dollars, so could only be reneted for price-affordability.Also keep in mind, if the subject matter is static, you can stitch a handful of low end DSLR images together and easily surpass 4x5 IQ. This represents the lowest cost, and highest IQ possible....
Then you have to do many many shots, and sure you can get a 1 gigapixel image, or greater. But it takes quite a long time. And you can also do panoramic stitching with large-format 4x5 scans, although to get the same reach as a 300mm in 4/3rds, you'd need a 2,250mm large-format lens (a little bit excessive), unless you want to do wider format panoramas.
Also, some subjects may be static enough for LF 4x5, but not static enough for the repositioning necessary to stitch an image. Then, you may need to do do focus-bracketing, to get the same effect as LF tilt. But looking at close-ups of LF scans vs. 1:1 of the output from my Oly E-3 (as much as I love it), I don't think that a 2 x 2 stitch (4 images total) from my E-3 will be as good as 1 large-format shot. Nor from other prosumer SLRs, like the Canons or Nikons.
Certainly, doing a 14 x 14 stitch with the E3 (assuming 10% overlap), that's 14 x 14 x 10Mp x 60% (due to overlap) = 1.176 Gigapixels. That will an incredible photo. IF the conditions were such that one had enough time to take it. I don't know if this can be done during photographic prime-time (sunsets, sunrises, dusk, dawn). This 1 gigapixel image took 13 minutes to photograph. I can only imagine how long it took to photograph this 17 gigapixel image took to photograph (presumably 17 times as long, so 3.7 hours).
See this google search.
Posting Permissions
Reply With Quote
Bookmarks