Sony 24-megapixel Imager Shows Up

"Tokyo, Japan—Sony Corporation today announced the development of a 35mm full size (diagonal: 43.3mm/Type 2.7) 24.81 effective megapixel, ultra-high speed high image quality CMOS image sensor designed to meet the increasing requirement for rapid image capture and advanced picture quality within digital SLR cameras...."

In a press release dated this morning, Sony Global has announced the new "full-frame" (i.e., 35mm-sized) imaging chipset presumably slated for the postulated "Nikon D3x" and "Sony A900." At 24.81 megapixels, the new CMOS chip is solidly within the range of what used to be the resolution territory of medium-format digital backs. Sony's press release emphasizes the "ultra responsive performance" and "broad dynamic range" and says, " Sony will target for mass production of this CMOS image sensor within this year."

_______________________

Mike  (Thanks to Adam McAnaney)

Featured Comment by Ctein: "So much misinformation, so little time [smile]. How many people test the claims they make or just repeat what they read somewhere? Me, I do a lot of testing (I get paid for it, after all). 

"1. Just about any good (not great) camera lens resolves over 100 lp/mm over a fair range of apertures. Really good lenses hit 200 lp/mm and higher. This sensor doesn't even come close to exceeding the capabilities of camera lenses.

"2. Pixel pitch ain't resolution. Assuming a Bayer filter array, the actual resolving capability of the camera will be about two-thirds the pixel pitch. With a slight crop, this sensor can produce a Super B print at 300 ppi; that will correspond to 200 lines per inch of actual resolution or 4 lp/mm. 4 lp/mm is a nice, sharp print, but it's nowhere near what viewers of even modest sensitivity can discern. Put that print between ones made with half the number of pixels and twice the number of pixels and said viewer will easily be able to sort them by sharpness. You may not feel the need of the extra resolution, but it will hardly be wasted on people who want it. 

"3. Do not confuse "need" with "usefulness." Few photographers need medium format over 35mm; even fewer need 4x5" over medium format. No one in their right mind would ever argue that the larger formats don't have their benefits or are a waste of time and money. Get the point?

"4. Unless you're obsessive about never, ever cropping photographs, basing your quality estimates on using the full 24-megapixel files is, at best, optimistic.

"5. Dynamic range in high-end DSLRs already exceeds by substantial margin what people got with color slide or B&W negative film. It rivals the extreme limits of color negative film. Sacrificing at most one stop of that range (and we don't even know if it's going to sacrifice that, because sensors are nowhere near their physical limits) is hardly a major loss of quality. 

"6. 75–150 MB files don't choke even my ancient Athlon 2400+ 32-bit machine with a whole 2 GB of RAM. A $1200 Macbook will just gobble them as snacks. This sensor won't tax anyone's computing or storage capabilities beyond affordability.

"Will most of you need a 24-megapixel camera? Absolutely not. Could most of you benefit from a 24-megapixel camera? Absolutely. 

"Will most of us be able to afford it? Highly unlikely [sigh]. (OK, on this point I am just speculating; no data here)."

Comments

[David Emerick]

I guess my a700 will be a backup camera next Fall!!

D

[Peter]

Bring on the rampant speculation and internet navel gazing!

Actually to add some content to this post and to link it to one of the blurbs on Photoborg, I'm very glad that a) Sony is going to be offering us a "full-frame" camera, and b) waiting for the next generation, in order to pick up an A900 on the cheap is going to feel like an eternity now.

I wonder what the pricing will be like.

[Dickon Whitehead]

How about a circular sensor!

[Stefan]

Nice to see more competition, but why is "full frame" still limited to 24x36mm (a ratio of 2 to 3)?
In the digital age multiple ratios would be possible within a circle of 43.3mm diameter:
3 to 4 = 25.9x34.6mm
6 to 7 = 28.1x32.8mm
1 to 1 = 30.6x30.6mm
Next to being very cool, a square would also have about 8% more real estate...

[Mike Johnston]

"Next to being very cool, a square would also have about 8% more real estate..."

...And would require significant redesign of the cameras, which apparently no one is interested in undertaking. Olympus is the only company which has so far shown much interest in rethinking basic camera design, although its efforts have pretty consistently gotten clobbered in the marketplace. Everything else is still pretty much legacy design from 35mm.

A square sensor would make much more sense, but don't count on it.

Mike J.

[Carl Leonardi]

Unless you have the best lenses in the world like the new Canon 200mm f/2 for only $6,000 or intend to make prints larger then 16X20, this will only give you very large files in the major over kill class. All the equipment needed to deal with these file sizes will be very expensive. At least 10 times more then a 40D or D300 class camera, lens and computer system. There won't be a zoom lens made that can match this resolution, so we may be looking at a come back for really good fixed lenses.

[Chris Crawford]

Carl Leonardi,

Computers will handle the bigger files just fine. 24mp is still a fairly small file. I scan medium format film with a Nikon LS-8000ED film scanner. The 16bit TIFFs it generates are around 350mb for a color image, and when I add a number of adjustment layers to adjust overall contrast and color as well as dodging and burning, I have a file that weighs in at 2-3GB. Yes Gigabytes. I keep these layered files as archives.

My computer is ancient by today's standards. A Powermac G4 1.25 ghz dual processor with 2gb of RAM. That computer is almost 5 years old! It handles these files just fine and is not slow. I can't imagine how much faster a new system would be. If my ancient Mac is fast with TRULY big file, I imagine that nobody will have trouble with the 24mp camera's files.

[Mark Roberts]

OK, it's early morning so my math skills haven't completely woken up yet, but...
The Sony chip puts 6236 pixels along the long axis (36mm) of the frame, giving you 173.2 pixels per millimeter. That means the Nyquist cutoff frequency will be half that or about 86.6 lpm.

There may not be many lenses that can meet that spec under real world conditions but there aren't as few as some are implying.

[Al Benas]

Two thoughts:
1. I'd love a square format; miss my old Hasselblad and Rollei. But I don't relish the probable $10k price tag.
2. I've been concerned about the growth in file sizes with this new monster chip, but I think, in reality, it will only effect the RAW file size (abt 2x). The final tiff size will be the same size, but all real pixels, no up-rezing required :)

[threadster]

It would seem the D3x is well along in development as it appears Thomas D. Mangelsen may be using the D3x currently.

I was at one of his gallery stores and I asked about his camera equipment and she mentioned the D3x. I said you mean the D3? Nope, D3x.

[Robert Roaldi]

At some point, the buying frenzy should level off, I'd think. In a previous blog entry, Mike referred to equipment having reached the "point of sufficiency". Each person's sufficiency is different, of course, and mine has probably been met at 10 mpix. My 5 mpix E-1 matches the ability of my previous 35 mm film gear to put prints on the wall of 8x10 size. My 10 mpix camera does even better. I think I have reached the point where I will stop buying for reasons of equipment improvement.

Of course, there are many people with more money than I and with more exigent needs than I, but the manufacturers must be chasing after fewer and fewer buyers. At least that's my uninformed guess. But then, maybe the 20 mpix FX form factor will be commonplace in 5 years, as much as 10 mpix is now. At that point, there may not be any 10 mpix cams made anymore, because no one will want them, or because they won't be profitable to sell at the prices that people will be willing to pay for such lowly equipment.

It's amazing, really, since few people print that big.

[Jason]

I'm just starting to wonder if as Einstein said, the tech isn't surpassing our ability to use it.

This is great for that tiny group that needs it, the small group that can afford it, the big group that wishes for it and the huge group that will bash your pics for not having been shot with it.

For all my geekery I've been having more fun photographically with my Bessa and two primes than anytime since I went 99% digital 4 years ago. Though my old go-anywhere, shoot anything 2 (two) MP Canon Powershot A60 came close. I still have great 8x10's on my wall from that cam.

[XebastYan]

I appreciate that Sony is going to produce this CMOS - Full Frame Sensor. I do this because i think that if Sony do it, DSLR's with Full Frame will have a decline in price.

My wishes are:
> Compact Full Frame DSLR's with non built-in batterygrips for under 2000€ or better 1500€
and ( i am romantic)
> a Full Frame Pentax DSLR where i can fit all the "old" K-Mount Lenses without the APS-C Crop

My Opinion:
I think that the Technical Construction-Plans for 35mm (Full Frame)Film Lenses aren't lost at the manufacturers, maybe the assembly lines are still work....
So if they change the Coatings of some Lenses from the "old Times" it will not all to be this difficult to develop the lenses for Digital 24MP- Full Frame & that we will become good performers on a Full Frame DSLR to "moderate" prices.

[Ed Hawco]

How about a plain old full frame sensor in the 12 MP range? It would be fast, sensitive, and practical. Why is that so difficult? Why so few, and so expensive?

[Ken Sky]

Hurray for all of use who kept our Minolta lenses and kept buying full frame lenses in the A mount even as the major off brand lens makers showed reluctance to support this lens mount. Perhaps now we'll see more competition from lens manufacturers. In any case, recent developments and releases by not just Sony but Nikon, Pentax and Olympus have broken the Canon hegemony. Competition can only breed cheaper and better equipment in the long run. We the consumers should applaud all the manufacturers because we're the winners.

[Jonathan]

'Sony's press release emphasizes the [...] "broad dynamic range"'

Well, that's the real question: is it possible (with the current state of the art) to pack this many pixels into a sensor and get dynamic range anywhere near as capable as the D3? Highly unlikely.

As capable as the D300? Maybe, but I suspect even the D300 will equal or slightly exceed the DR potential of this sensor.

For the artfully-inclined amateur on a medium-to-high budget, the D300 and D3 have already set the bar as high as it needs to go. We're at the point where resolution is quite sufficient for fine prints at gallery sizes, and now we just want better dynamic range and finer microcontrast within that resolution.

Unless you're printing massive murals meant to be viewed at close range, the D3X is just going to be excess resolution you never print.

I can see this being a hit for professional sports shooting and photojournalism - e.g. shooting a basketball game with a lens wide enough to encompass half of the court, and just cropping out 12mp close-ups later. Or the same approach in a chaotic news event - shoot now, compose later.

But for someone doing their framing & composition on-the-spot, 24mp is expensive overkill.

[Ken Tanaka]

It will be interesting to see how this new sensor performs. But I believe that most amateur, and amateur pro, photographers will discover that 25 Mp is far more image data than they need...or even want. Worse, it won't produce better no-light baby and travel snaps.

But it sure will sell hard disks.

[Peter Popp]

I think everybody wins with this, and I'm a Canon user who will probably never own a full-frame camera.

[Feli]

Interesting, it looks like it is 12bit, not 14bit or 16bit.

But it is very fast at 6.3 frame/s (12bit).
I'm guessing that's why they didn't up the bit rate.

Feli

[Charlie H]

It will be intersting to see where the full frame vs. aps sized sensor wars finish up. At today's prices for Canon's 5d, it makes you wonder whether aps-c is a passing phase. I would hate to see that just becuase it would be incredibly wasteful to have all those DA, fx etc. lenses become useless or obsolete. But I think its basically inevitable. When you think about it, it makes great sense from a manufacturers perspective. Discontinue your old lenses, buld new ones, then rebuild your old ones. ch

[John Camp]

"Nice to see more competition, but why is "full frame" still limited to 24x36mm (a ratio of 2 to 3)?
In the digital age multiple ratios would be possible within a circle of 43.3mm diameter"

This is exactly what I expect Leica to do with the new R cameras. There have been a number of occasions when, in discussing future products, Leica people have said the R10 will be "full frame or larger." There's no way they would throw all their legacy lenses over the side, so I believe this means that they will come up with an aspect ratio that will provide more territory, but still work with their standard 35mm lenses. Leica desperately needs something that would allow them to claim the high ground, given the cost of their system. A square is unlikely; nobody but the aesthetically impaired would wish for a square (wait...wait...that's just a troll, you can ignore it.) In any case, I think a slight rectangle, somewhere between 3x4 and 6x7, is likely. I could be wrong; I often am.

JC

[Dennis]

Ahhh, the old square & circle arguments :)

> In the digital age multiple ratios would > be possible within a circle of 43.3mm
> diameter:
> 3 to 4 = 25.9x34.6mm
> 6 to 7 = 28.1x32.8mm
> 1 to 1 = 30.6x30.6mm
> Next to being very cool, a square would
> also have about 8% more real estate...

But existing lenses for 35mm systems don't project 43mm diameter circles; they project 24mm x 36mm rectangular images (in some cases). A 30mm square is taller and would require a taller mirror that would not have room to flip up without smacking into the back of some lenses (or the lens mount itself).

24x36mm is the most real estate you can put into a camera that uses the existing 35mm lens mounts.

Ultimately, the best sensor aspect ratio is the one that most closely matches your desired file ... if you're going to crop regularly to rectangular, why shoot square requiring a taller than necessary mirror (further distance to sensor and extra vibration to deal with) ... until we get the EVIL camera, anyway.

[Marc Rochkind]

The significance of this new sensor (and the new Canon 1Ds III) is that the price of MF digital resolution (with smaller pixels and probably more noise) is going to drop by around 2/3 within a year. (I'm guessing that instead of $30K for a MF cameras with digital back, a camera with this sensor will go for under $10K.) And, that will drop more over the next 5 years or so, whereas MF digital back costs won't drop nearly as much because the production numbers are so low.

[Scott]

1. Good job Sony for expanding the full-frame sensor line another notch!

2. This gives an argument to switch to a Mac G4 with 2 Megs of RAM from my current Toshiba laptop running Vista, same RAM, which frankly doesn't do too well with large files.

3. Where am I going to put all these huge image files now?

[Jeff Hartge]

Someone spot me a clue...

It would seem that there would be two approaches to building sensors:
1) build a "low number" of "high quality" sensor sites on a senor (the current approach).
2) build a ludicrously high number of "low" quality sites on a sensor.

The second case would be more analogous to "film" - where the "quality" of the individual light sensitive molecules is highly varied but is good on an average basis.

It would seem that the second case would be easier to build. They already have these near ludicrous densities for digicams (and they still have some individual pixel quality standard). They would just need to do it over a larger area using lower quality standards (who cares if that site is too big, too small, doesn't work at all, two reds next to each other, etc.). All that matters is that the average response within a given region of the chip is consistent.

In the first approach, if one site out of 12 million fails to respond, the chip needs to be thrown out! In the second case, so what if 10% of the sites don't work, there are enough sites in a given area to do the job.

Further, who cares if a site fails during the course of usage. Again, it just means one less site for averaging.

It would also eliminate the need for an anti-aliasing filter and would eliminate the need for the Bayer processing.

Am I misssing something?

[peter in bangkok]

I am buying a HOLGA

[michael]

How many people routinely print bigger than 8.5xll or 11x14? If you don't, you don't need more than 8MP.

I've said this before: instead of more diodes, how about latitude/dynamic range of film? That's what needs fixing.

And you're going to need better lenses than now.

[Kelvin L.]

Hooray! I'll be getting poorly composed photos with dull subject matter in 24 megapixels instead of 11! ;)

[Dennis Allshouse]

Way to go Chris! I'm using a similar vintage G4 laptop which I think will be underpowered(note Chris is using a dual processor G5)

Anyway the native print size at 300 dpi would be roughly 13 x 20, is that big enough!? It seeming likely that the camera will also be maxed out in the bells and whistles department ala the D3 or D2x and hence very expensive. I would love to see this chip in a D300 package or even a 16 meg chip.

[Adam Zolkover]

I can't say that I didn't have some pangs of yearning when I saw this. But I figure it this way: megapixel-wise, the 10mp K10D is about as much as I need. For me, the reason to go to full frame would be increased high-ISO performance, and better dynamic range. And I just don't see that happening with that pixel density. What I'd really like is a 10 to 15mp full-frame (or near full-frame) sensor in a k-mount body approximately the size and shape of my old Pentax MX.

[Sam Murphy]

I wonder if anyone has the guts to go the other direction with megapixels, ie. a 3MP full frame dslr for newspaper reportage? ISO 409,600 anyone?

[hugh crawford]

A few observations about the 35mm dslr state of things

Anything more than about 12 megapixels is wasted if you are shooting hand held unless you are using high speed strobes or very high shutter speeds. Cheap studio strobes are not "high speed" btw

The existing lens systems aren't quite up to the task already , for instance Canon can't seem to figure out how to make a decent wide angle lens, which took a lot of the fun out of their full frame cameras

The open loop phase detection autofocus systems aren't up to the task of focusing a lens at f5.6 for a 20x30 print. I think that I read that canon specs their autofocussing to be good enough for a 5x7 print at f8 or something like that. Live view contrast detection focusing will do better, but will be slower.

DSLRs can give you a better image than 99 percent of the non tech-pan using 35mm photographers ever got. I personly hated 35mm cameras for anything other than jobs where image quality wasn't the point, and for most professional work image quality is not at the top of the list of "needs" or even "wants"

I don't really need a better digital canon/nikon/leica , I need a digital Yashicamat, or a Mamiya press 23.

Too bad that Sony won't be putting that new sensor into a camera along the lines of their R1

[Nick]

Yes, yes, very nice, but my 5D has all the resolution I need and with 10000 pics a year it should last me another 9 years... So, Sony, how about putting all that effort in a compact, fixed lens, aps sensor equipped camera like the Sigma DP-1 is/was supposed to be... That way my 5D will last me even longer!

[Ctein]

Dear folks,

So much misinformation, so little time [smile]. How many people test the claims they make or just repeat what they read somewhere. Me, I do a lot of testing (I get paid for it, after all).

1) Just about any good (not great) camera lens resolves over 100 lp/mm over a fair range of apertures. Really good lenses hit 200 lp/mm and higher. This sensor doesn't even come close to exceeding the capabilities of camera lenses.

2) Pixel pitch ain't resolution. Assuming a Bayer filter array, the actual resolving capability of the camera will be about two-thirds the pixel pitch. With a slight crop, this sensor can produce a Super B print at 300 ppi; that will correspond to 200 lines per inch of actual resolution or 4 lp/mm. 4 lp/mm is a nice, sharp print, but it's nowhere near what viewers of even modest sensitivity can discern. Put that print between ones made with half the number of pixels and twice the number of pixels and said viewer will easily be able to sort them by sharpness. You may not feel the need the extra resolution, but it will hardly be wasted on people who want it.

3) Do not confuse "need" with "useful. " Few photographers need medium format over 35 mm; even fewer need 4 by 5 over medium format. No one in their right mind would ever argue that the larger formats don't have their benefits or are a waste of time and money. Get the point?

4) Unless you're obsessive about never, ever cropping photographs, basing your quality estimates on using the full 24 megapixel files is, at best, optimistic.

5) Dynamic range in high-end DSLR's already exceeds by substantial margin what people got with color slide or B&W negative film. It rivals the extreme limits of color negative film. Sacrificing at most one stop of that range (and we don't even know if it's going to sacrifice that, because sensors are nowhere near their physical limits) is hardly a major loss of quality.

6) 75-150 MB files don't choke even my ancient Athlon 2400+ 32-bit machine with a whole 2 GB of RAM. A $1200 Macbook will just gobble them as snacks. This sensor won't tax anyone's computing or storage capabilities beyond affordability.

Will most of you *need* a 24 megapixel camera? Absolutely not. Could most of you benefit from a 24-megapixel camera? Absolutely.

Will most of us be able to afford it? Highly unlikely [sigh]. (OK, on this point I AM just speculating, no data here).

pax / Ctein
[[ Please excuse any word-salad. ViaVoice in training! ]]
=========================================
-- Ctein's Online Gallery http://www.ctein.com
-- Digital Restorations http://photo-repair.com
=========================================

[John Roberts]

There seems to be a substantial number of "amateurs" (including me) who believe 6-10 MP is plenty for what we do with a camera. This group also seems to be wishing for camera manufacturers to focus on increasing the dynamic range of a 6-10 MP sensor instead of forever trying to cram more pixels on the same size piece of real estate. Maybe one day they'll start listening to our segment of the market. Until then, I'll just keep shooting with my E-500 until it wears out.

[Feli]

>"5. Dynamic range in high-end DSLRs already exceeds >by substantial margin what people got with color >slide or B&W negative film. It rivals the extreme >limits of color negative film. Sacrificing at most >one stop of that range (and we don't even know if >it's going to sacrifice that, because sensors are >nowhere near their physical limits) is hardly a >major loss of quality.

I'm sorry, but I really do not agree with this and don't believe this information to be correct.

Digital may deliver images with less noise, but still has a ways to go in terms of dynamic range.

Many cameras may benchmark to 11 or 12 stops, but really only deliver a usable 8-9 stops, with the Fuji S5 Pro being the exception at 10 stops.

There may be some high end medium format backs that deliver better performance but at a cost of tens of thousands of dollars, they aren't exactly mainstream.

Slide film gets about 6-7 stops, so digital certainly has it beat.

Both color and black and white negative can easily capture 10-14 stops of range.

Also while DSLR may deliver higher color accuracy than film, they have until recently been mostly limited to 12bit, which results in poorer tonality than negative film.

Cheers,

Feli

[ggl]

I've seen all the useless math and read most of the "theories". But the fact is that images from a full frame digital camera look way better than images from cameras with smaller sensors (read APS-C, DX, you name it). Just compare an image from a 5D with the same image from a D300 and you'll see what I mean. It simply looks better (you can even use the same lens via a Nikon->EOS adapter). I'm not only talking about resolution only; it also boils down to cleaner images, higher dynamic range, smoother tones. The difference is similar with 35mm film vs. medium format (say 6x4.5). Anyway, digital 35mm full frame cameras are still too big (yes, even the 5D). I wouldn't use one for shooting street photography for instance.

[Rubén Osuna Guerrero]

It was considered 40 lp/mm of real detail as the key mark for 35mm format and moderate prints. In order to resolve that real detail you need more than 80 lp/mm of sampling frequency (Nyquist theorem). The berfore mentioned 80 lp/mm leads to 6,25 pixel spacing (not real pixel size...). Then, 24 MP isn't too much at all for any 35mm lens. The limit resolving power of a lens at low contrast levels isn't relevant. The 40 lp/mm mark is far more interesting in practical terms. This frequency of the signal (real detail) can be transmited at 50% contrast levels or more at the axis in many good 35mm lenses.

[Andrew]

http://www.dpreview.com/news/0801/08013109sonypma.asp

So the Sony A900 will be available in 2008, with a 24.6MP CMOS FF sensor & the camera has incamera AS (SSS)

My understanding is this is not going to be a pro body, but a competitor to the Canon 5D.

Things are about to get intersting :)

[Ctein]

Dear Feli,

As one who has been printing all kinds of films, all different ways, for 40 years, I'd argue you're not really comparing apples and apples.

You do not get 14 stops of "usable" range out of either B&W or color neg film without very abnormal processing (in which case you can push it to 20 stops). Even then, there's a serious question of how usable it is. Have you ever tried pulling a decent print from even a 10-12 stop range negative? It takes fairly heroic measures (you can just print on really low contrast paper, but the results look like crap).

From a practical viewpoint, trying to print more than 10 stops of subject range is usually not workable nor worth the effort. I've only bothered a coupla-handful of times over my entire career, and I go after really long-range subjects with gusto.

If one applies that comparable level of effort to digital files, one can get the benhmarked dynamic ranges. I agree that routinely the cameras deliver a few stops less than the benchmarks... but routinely processing and printing color or B&W negatives never gets you near even 10 stops.

I've not seen evidence that the tonality of 12-bit digital files is worse than 35mm or even medium format negative film. But I'm willing to be convinced. Can you point me to a reference for that? Comparison images would be nice, but I can be swayed by cogently constructed argument. Thanks.

pax / Ctein

[Ctein]

Dear John,

I'm pretty much in your camp. I'm happy with my 6 MP Fuji Finepix S6000 for anything but my "most serious" photography. Makes perfectly decent 8x10's. Not that 24 MPixels wouldn't be a lot nicer, but I ain't gonna be able to afford that puppy. And at $300, it'd be hard to complain about the Fuji.

'Course I will. Dynamic range is, at best 8-9 stops. Better than I'd get from slide film, not as good as I'd get from B&W negs, nowhere near as good as I get from color neg (my mainstay). And, as has been discussed frequently here, digital cameras don't fail anywhere as gracefully as film when they hit the dynamic wall, so exposure is tres finicky.

Unfortunately, I don't see the manufacturers catering to the likes of thee and me. There's an assumption that low-pixel count cameras are for duffers who don't care as much about image quality. I fear the truth is that likely DOES represent most of the market.

We get the benefits of technology trickle-down: my Fuji with its 1/5th scale sensor runs rings around most full-frame sensor cameras of a decade ago. But I believe that's all we'll ever get. In the film business, I'd opine that technological improvements were fundamentally amateur-market driven. In the digital, it's definitely pro-driven.

I'd love it for someone to prove me wrong on this one.

pax / Ctein

[Feli di Giorgio]

Hey Ctein--

I'm a compositor and visual effects supervisor for film and television and have worked with an ungodly amount of scanned film, digital capture and output over the past 15 years.

I'm also a documentary photographer, who predominately shoots black and white (Tri-X), but of course I have also shot digital. You can see some of my work at my website www.elanphotos.com

I certainly think the information is there on the negative and I agree that it far exceeds what the final output media (paper print or release negative for theaters) can display.

But I think we can agree, that as a starting point it is desirable to have as much information as one can capture. As you know, if you don't you'll never be able to recover those blow highlights or dig into shadows etc. This is why I never liked slide film.

As you said, trying to fit the entire range of a negative into the range of traditional silverprint paper is pretty close to impossible using analog methods and as you say a total nightmare to execute. It is even more difficult if the film was specially processed to stretch it's range to the limits.

But this is where digital comes to the rescue. My workflow is hybrid analog / digital.

I capture analog (Tri-X), scan it at 16bit and then process it in Photoshop or a program called NUKE, that was developed by a company I worked for called Digital Domain. It is now distributed by The Foundry in the UK. If the range of the negative exceeds what the scanner can capture in one pass, you can wedge your scans and then assemble a single image with expanded HDR.

Obviously with digital you can quite easily compress the data captured from the negative into the more limited range provided by your print media.

This file is then sent to a Rhino LVT, where a new negative is generated, which is then wet printed onto fiber paper.

Or the digital file is sent directly to a inkjet printer or something like a Fuji Lightjet for final output.

"I've not seen evidence that the tonality of 12-bit digital files is worse than 35mm or even medium format negative film. But I'm willing to be convinced. Can you point me to a reference for that? Comparison images would be nice, but I can be swayed by cogently constructed argument."

I scan all of my negatives on a Nikon 9000ED at 16bit or have them done on an Imacon and I see better tonality and range than the 12 bit output from several DSLR cameras I have used. Granted the film data is a lot noisier than digital.

At work I use images that come off a Arri laser scanner and without a doubt they have both more DR and color resolution than the vast majority of digital capture I see, although lately there have been some very impressive developments in that area.

A few weeks ago I saw a set of digital files that displayed a jaw dropping amount of data. Apparently these were straight captures, not assembled by a wedged exposure. Unfortunately I'm not sure where they came from, but they were not from a DSLR.

During my career we evolved from using 8bit (Apollo13 was done in 8bit) to 12bit and eventually 16 bit and sometimes even higher for elements that are generated in 3D packages. There is a noticeable advantage to the higher bit depth data, once you start to manipulate it.

Higher bit depth certainly makes a difference in digital capture. There was an epic thread on Fred Miranda's website comparing the tonality of a 12bit Canon to the 16bit output from the Leica DMR and the advantage of the higher bit depth was very obvious.

If I had to make a prediction about the future I would say that this will be a moot argument in 6-8 years.

Cheers,

Feli

[Ctein]

Dear Feli,

Oh yeah, we can now do amazing things with hybrid workflows!

Thanks for that info on bit depth. Given that the film scans are a lot noisier than the digital images, are you sure you're seeing the results of source bit depth and not other psychophysical effects?

Several years ago, I did a comparison photo session with Michael Reichmann, him with a very expensive medium format digital back and me with my film-using Fujica GA645. I noticed that his prints looked 'flat' compared to my darkroom prints-- they seems to lack subtle tonal differences-- gradation, if you will. I attributed that to the source.

Then I observed that digital prints I made from my 16-bit film scans had that same 'flat' look, compared to darkroom prints from the same color negs. These were prints that matched as close as I could get in overall tone and contrast, and there wasn't anything especially unusual about the negs-- straightforward printing jobs.

I saw this with several different makes and models of printers. I have a feeling that it's a problem in the workflow between Photoshop and the printer. I'm deeply suspicious of the printers' 8-bit drivers. But that's just a feeling; I don't really know where the problem is. Just pretty well convinced it wasn't the source material.

Are you sure you're entire workflow's maintaining the 16-bit (or at least 12-bit) depth, all the way through the output stage?

Your write-back-to-a-neg-and-print-in-the-darkroom approach sounds pretty convincing to me, if you're sure that the film writer's holding the full bit depth.

pax / Ctein

[Feli]

"Dear Feli, Oh yeah, we can now do amazing things with hybrid workflows!"

Yes, yes! In a way, digital has unlocked the full potential of what an analog capture media like film can deliver.

"Thanks for that info on bit depth. Given that the film scans are a lot noisier than the digital images, are you sure you're seeing the results of source bit depth and not other psychophysical effects?"

Maybe. As you know film and digital have a very different character when it comes to dealing with highlights and shadows. So, in some cases one will have an advantage over the other. Film certainly wins when it comes to dealing with highlights, unless you carefully expose digital for the whites and then push the shadows and mid tones up in the RAW processor. And that works really well, because digital has far less noise in the shadows than film. I'm dying to play with the new Nikon D3 for this very reason...

On a side note, a few years ago I got into a somewhat heated discussion with a man in a cafe, about the difference between how film and a CCD 'fail' in the highlights. Later on I discovered that I had just told Melvin Sokolsky quite bluntly that he didn't know what the hell he was talking about. In hindsight he took it pretty well. ;-)

"I saw this with several different makes and models of printers. I have a feeling that it's a problem in the workflow between Photoshop and the printer. I'm deeply suspicious of the printers' 8-bit drivers. But that's just a feeling; I don't really know where the problem is. Just pretty well convinced it wasn't the source material."

I have noticed the same thing. It could be the 8bit drivers, but I'm not sure. Does the 8bit driver limit us to 256 steps on the paper? I've always had a hunch that it was the ink and paper media itself, in particular with pigment based inks, which can look chalky. Perhaps it has to do with how the light scatters on the surface of such inks. I certainly have noticed a vast improvement over the past few years though, as the dmax of papers and inks has improved.

I also have a sneaking suspicion that sometimes it is the lack of micro contrast in the digital files. Anti-aliasing filters over the sensor tend to obliterate all that low level acutance and then with the lack of grain and truly sharp edges your eye no longer has something to lock on to and it starts to 'hunt'.

Several people have mentioned that prints from the M8 and Leica DMR have more bite and punch, because the files contain a lot of micro contrast. Both cameras lack an AA filter over the sensor.

I also think it has to do with the fact that the color correction tools in Photoshop are quite primitive and do not allow you to really fine tune an image to the extent that you can in something like Nuke. In Nuke for instance you have a color corrector that works like a multi-band equalizer, so you can make much finer adjustments than just high, mid, low. In addition these color correctors can be stacked or daisy chained. The range that they effect can then be offset and you can adjust the tonal range in ever finer increments. That way it is quite simple to adjust the micro contrast in an image at a very low level.

"Are you sure you're entire workflow's maintaining the 16-bit (or at least 12-bit) depth, all the way through the output stage?"

I know for certain that any image that is read into Nuke is automatically converted to 32-bit floating point data and maintained as such until you decide what format it is output to. All operations in the program such as color correctors and filters are executed in full 32bit.

Photoshop claims 16bit, so I'll take their word for it.

The files that go to the Lightjet are knocked down to 8bit. But that data is really only used to drive the lasers that are exposing the photographic paper. And that photographic paper produces an image that is continuous tone, unlike a inkjet device that can only 'draw' so many shades of a tone.

The Rhino LVT negative also produces a continuous tone image, when it is used in a traditional enlarger. I'm not sure what goes on inside the Rhino LVT, but you can give them a 16bit image. I just don't know if this is sampled down to 8bit internally. The Arri Laser recorder we use in my industry takes 10bit log or 16bit linear files and those new negs can be printed up and down like a real one, so maybe the LVT works in a similar fashion.

Cheers,

Feli

[Barbu]

Whoa! I'd pay good money just to see a discussion or a test session between Ctein and Feli di Giorgio!

[Keith Forbis]

I think this is good news, and I'm glad that there is still some excitement about what is next with digital.

Personally, I prefer the 2/3's ratio. Sometimes I crop to a different ratio, but not usually.

Back when most people were still using film, the best APS camera/lens/film combo couldn't match the quality of good 35mm. And the best 35mm camera/lens/film combo couldn't match the quality of good medium format, etc. Well today the same is true of APS sized sensors vs. full frame vs. medium format. And of course, some people dealt with a lot more hassle because they wanted to shoot 16X20 instead of 4X5. I don't think these choices are a bad thing, and I'm still going to have a lot more images in a lot less space then when I shot film.

Chris Crawford,
"2-3GB?" Egad!!! I haven't worked on any scanned medium format negs yet, my system definitely wouldn't be able to handle 2-3GB. I have gotten close to 1GB and it gets veeeery ssslllloooowwwwww.

Ed Hawko mentioned a full-frame sensor in the 12MP range. There is the Canon 5D, but how about something more like (and priced like) a 10/20/30/40D but with full frame?

Charlie H,
I think APS sized sensors will stick around because they provide excellent quality at a reduced price (compared to full-frame). Just about everything about an APS sized SLR can be smaller and therefore less expensive, it's not just about the sensor expense.

Hugh Crawford,
I think that when the digital thing really sort of plateaus, then other companies will start making niche cameras, and I think that should be pretty interesting.

Feli,
while your conversation with Ctein is a bit out of my league, I think I know what you mean about micro contrast. It is something that I have been becoming much more aware of and unhappy with. I'm not familiar with Nuke, but somebody just told me about ACDSee Pro and how there are "equalization sliders." I downloaded the free trial version and have just begun experimenting with it. The only reason that I'm interested in ACDSee is for those sliders, and so far I'm finding it very interesting. Photoshop curves can be a wonderful thing, but I find the "equalization sliders" to be much more of what I'm looking for when it comes to fine/subtle selective tone adjustments. Maybe it is just a more intuitive way for me to work, or maybe it is just a better tool... I don't know.

[Ctein]

Dear Feli,

Regarding the gradation problems we're both seeing in prints, I don't think it's the medium itself, because my semi-glossy inkjet prints have a density range that isn't much different than my chromogenic darkroom prints, plus the color gamut is better. The darkroom prints are still conveying subtleties of tone and color that prints made from high-quality scans of the same negatives don't. My workflow is entirely 16-bit until the output stage, but I know that the drivers today are all 8-bit (except for some Unix stuff where people rolled their own 16-bit ones).

You could be right about there being additional tonal problems with digital camera files, but I don't have any way to test for this because I don't know how to get output is that isn't degraded.

Someone (not me) might try taking a known high-quality digital file and using GIMP to print it out under Windows and under Unix (using said 16 bit printer drivers). It might not prove anything, but you never know ...

The problem may lie in the color management and profiling. All that stuff is still firmly 8-bit. Since you're warping the color space when you profile, you're going to get effects analogous to picket fence histograms. So maybe this isn't a printer or printer/driver problem but a color management deficiency? I can't think of any way to test for that until we get 16-bit color management a few years down the line.

Being limited to 256 grey levels, per se, shouldn't be producing these obvious problems. The human eye's got lousy tonal discrimination. 256 ain't visually perfect, but it's getting close.

Regarding your ideas about 12-bit vs 16-bit camera files ... it shouldn't be too hard for a code jockey to write a little utility that would take a 16-bit raw file and round off all the data to 12 bits. You could then take both versions of that file into your image processing program, process them with identical settings, and print them out. If the un-rounded version showed better tonal qualities, that would prove for certain that 16 bits was meaningfully better than 12.

I wasn't aware of Nuke. Thanks for bringing it up! I just downloaded the file. I'll have to play around with it some.

One can do extremely subtle tone and color manipulations in Photoshop by using 16-bit adjustment layers. You can make small adjustments in layers and then dial back the opacity. For example, in curves, moving a tone from 128 to 129 would be the finest adjustment you could make. But if you set the layer for 1% blend, you can actually make an adjustment from 128 to 129.01.

Not saying that Nuke doesn't do a lot better (as I haven't tried it yet ), but I just thought I'd mention this hack.

pax / Ctein
[[ Please excuse any word-salad. ViaVoice in training! ]]
=========================================
-- Ctein's Online Gallery http://www.ctein.com
-- Digital Restorations http://photo-repair.com
=========================================

[Feli di Giorgio]

You'll like Nuke. It wasn't designed for print work, but it is astonishing what you can do with it. Sooner or later they will get icc profiles for viewing in there. Right now I have to final check everything in Photoshop.

One great thing about Nuke is that it is scanline based and only reads the entire image into memory at once, if it has to perform an operation on the entire image (rotate etc). In any case you can assemble incredibly complex composites with a reasonable amount of memory.

[expiring_frog]

"I know for certain that any image that is read into Nuke is automatically converted to 32-bit floating point data and maintained as such until you decide what format it is output to. All operations in the program such as color correctors and filters are executed in full 32bit."

I'm sure most of you know this already, but I thought I should point out, just in case, that 32-bit floating-point is a very different beast from a 32-bit integer, so a comparison based on bit-depths alone is not quite justified. Assuming the IEEE standard, you get precision equivalent to a 23-bit integer for the upper half of the linear luminance range (i.e. the first f-stop), double that for the next quarter, and so on. The expose-to-the-right paradigm goes out of the window if you could somehow work entirely in floating-point starting from the A-to-D conversion, because every f-stop is recorded with (roughly) the same precision (unlike a linear 8, 16 or 32-bit integer encoding). I would be interested in knowing whether, in the same vein, a 16-bit floating-point representation is more "natural" (given the eye's response characteristics) than 16-bit integer.

(Am ignoring gamma correction here.)

[Roger S]

I think this thread has a similar example of a 5D vs 6x7 where the film has more color speckles that make it look less flat. I just wonder if not suppressing digital noise wouldn't help:

http://photo.net/bboard/q-and-a-fetch-msg?msg_id=00Nzyh