Best camera setup for Fine Art Repro? Scanning backs? High MegaPixel DSLR?

[EricWHiss]

Multishot is better, and reproduction is the number one use case.

I've shot some reproduction jobs of very well known artwork with both my AFi-ii 12 and my CF 528 using the same exact lens, 90mm APO Rollei Schneider Macro.
I will not be allowed to share these images in any form so please don't ask, however my conclusion was that they are very close, however for sheer image quality the CF 528 was better. There are many facets to consider, color, detail, tonality, texture etc.   In micro step mode the CF 528 has a higher resolution file and at true color (non interpolated).  I had to run the Leaf files through a two step sharpening - first small amounts in C1, then again in LR4 with the detail slider up high - just to get the fine detail close between them.  There's probably some math that shows what the resolution advantage true color has over bayer interpolated files, but needless to say its there.  It's also there for lower contrast detail too - like fingerprints on scotch tape - single shot backs of the same resolution completely miss that stuff when it gets small enough, however multishots do pick that up.    It's quite astonishing.   This can be significant when you are trying to capture the brush work and other subtle nuances.  With regard to color - that's tricky. I could see a case where one could argue either is better but out of the box for paintings and stuff, the Leaf and C1 is probably better than the CF 528 and Phocus in Reproduction mode, but there are noticeable differences.  Obviously much less when you make your own profiles.  Separation of color in small regions is better with multishot and tonality seems more real.    But the work flow advantage of a single shot file is really quite significant. It's not just the time, but also the conditions required for micro step - even lighting, vibration free.  Mostly I use the AFi-ii 12 when I can because its much faster and for many things quite close.  With regard to color - that's tricky. I could see a case where one could argue either is better but out of the box for paintings and stuff, the Leaf and C1 is appears to be better match to original than the CF 528 and Phocus in Reproduction mode, but there are noticeable differences in color between the two set ups. But when you make your own profiles they get much closer. 

[Theodoros]

Obviously, if you have friends in a well-funded museum who are ready to pay you $1000 per image to do an exhibition catalogue, the price of equipment is not going to be an issue

Edmund

Edmund

1000 per capture Edmund? Are you sure?  That means if one does 20 paintings ...a 20K day income?    Where is this museum? Can you send me an address of it? 

[torger]

Yes single-shot is a great workflow advantage, that's one major reason I think people use them when it's become "good enough", and now 135 systems are good enough for some applications which can be a further workflow improvement although that is more debatable.

Concerning color on large fields the single-shot should not be worse than the multishot as the single-shot get sufficient sampling. But in pixelpeep the multishot has a clear advantage, especially if it's moving sub-pixel. Bayer singleshot demosacing is in fact a lot of guesswork, it's about making a pleasing and quite likely result, accuracy is not possible to achieve.

Here's a nice comparison between single shot 80 megapixels and 4-shot(?) 50 megapixel. Note that Hassy also have the 200MS version which is 6-shot subpixel. Now Hassy bases their multishot cameras on the CMOS sensor and the lower noise of that should further improve the multi-shot quality. I recently saw a H4D-50MS on ebay for £10k by the way.

https://captureintegration.com/leaf-aptus-ii-12-hasselblad-4d-50ms/

The worst problem as I see it with the single shot is that they create detail that's not there and false color on fine structured details. You can see it in this example crop, quite subtle but the grey textile has some colors in it in the single-shot which is not there in the real object, and the multishot captures that correctly. You can also see that the magenta color area does not match either but I don't know the reason for that.

https://captureintegration.com/wp-content/uploads/2013/04/Screen-Shot-2013-04-15-at-4.14.37-PM.png

[Theodoros]

I could see a case where one could argue either is better but out of the box for paintings and stuff, the Leaf and C1 is appears to be better match to original than the CF 528 and Phocus in Reproduction mode, but there are noticeable differences in color between the two set ups. But when you make your own profiles they get much closer. 

I don't see how can one use "out of the box" colour profiling for paintings Erik... It's a part of the job to achieve perfect profile calibration and 528c is perfect when you do... No interpolated colour can be better than "perfect true colour" ...can it?

[torger]

More on single vs multishot using a bit older gear:

http://www.luminous-landscape.com/images-103/Figure-2-full.jpg

From the review http://www.luminous-landscape.com/reviews/cameras/h3d50ii.shtml

http://www.luminous-landscape.com/reviews/cameras/h3d50ii.shtml

However that review also points out Quote "Based on the color “blotchiness” and the lack of resolution in single shot mode, I would not consider using this camera in anything other than Multishot mode. I hardly need to say this, but the PhaseOne System [P65+] is far superior to the H3D 50 II in single shot mode, as it never shows any “blotchiness” and has much higher resolution." End quote.

A comment on that is as the P65+ has microlenses it should show a bit less aliasing (samples more of the 6x6um square), but I would suggest that the difference seen there is more due to luck than other, as moire can disappear with slight defocusing, and perhaps Phase One bias, suggested by that 60 megapixels is considered "much higher resolution" than 50... Mark Dubovoy has had some strange things to say about dynamic range too so I would not take his words in that review too seriously.

[Bart_van_der_Wolf]

No interpolated colour can be better than "perfect true colour" ...can it?

Nobody claimed that a Bayer CFA image is 'better' than a micro-step (or a color-wheel with a monochrome sensor) capture.

Tri-chromatic (R/G/B) color capture is still a relatively rough approximation of the full visible spectrum reflectance ...
In that sense, even micro-step sensor capture is also 'interpolated' color, although more predictable than from a single Bayer CFA. It does require extremely constant continuous lighting, where temperature and long exposure time are enemies of delicate artwork.

Of course, although it produces a more involved workflow, stitching with down-sampling can produce even better results. For recurring jobs, it will pay off to use a contraption like Ted describes here. One of the immediate benefits is the higher level of control over the lighting of the artwork, because only a part of the image needs to be lit and lighting angles are constant for that relatively small area of the total surface.

Cheers,
Bart

[Theodoros]

Yes single-shot is a great workflow advantage, that's one major reason I think people use them when it's become "good enough", and now 135 systems are good enough for some applications which can be a further workflow improvement although that is more debatable.

Concerning color on large fields the single-shot should not be worse than the multishot as the single-shot get sufficient sampling. But in pixelpeep the multishot has a clear advantage, especially if it's moving sub-pixel. Bayer singleshot demosacing is in fact a lot of guesswork, it's about making a pleasing and quite likely result, accuracy is not possible to achieve.

Here's a nice comparison between single shot 80 megapixels and 4-shot(?) 50 megapixel. Note that Hassy also have the 200MS version which is 6-shot subpixel. Now Hassy bases their multishot cameras on the CMOS sensor and the lower noise of that should further improve the multi-shot quality. I recently saw a H4D-50MS on ebay for £10k by the way.

https://captureintegration.com/leaf-aptus-ii-12-hasselblad-4d-50ms/

The worst problem as I see it with the single shot is that they create detail that's not there and false color on fine structured details. You can see it in this example crop, quite subtle but the grey textile has some colors in it in the single-shot which is not there in the real object, and the multishot captures that correctly. You can also see that the magenta color area does not match either but I don't know the reason for that.

https://captureintegration.com/wp-content/uploads/2013/04/Screen-Shot-2013-04-15-at-4.14.37-PM.png

Thanks for posting this comparison... the advantage of "true colour" capture in both resolution and colour is pretty much obvious and this is only a 4x capture....

By the way, Hasselblad's 200MS uses the 4x true colour result as captured by 50MS and then it interpolates that colour to apply it on the rest of the resolution that it creates, while 528c/22MS and Sinarback 54H provide true colour capture in their full resolution 16x sub-pixel mode (Sinar's eXact too).

[Kolor-Pikker]

I recently bought a 645Z and do all my art repro with the FA 120mm Macro now, I used to stitch shots with a 5D before! But as far as printing on 44" canvas goes... it's not a particularly demanding medium in terms of resolution, nor does it have a particularly wide gamut, so a single shot is pretty much enough to reproduce a meter-wide painting in crisp detail, and going any bigger I would probably stitch two-three shots.

I think 80MP backs, scanning backs, and multi-shot systems may be nice for some real big artworks or getting that last 5% of image quality, but I tend to stop at the point where I can see bristle indents in the brush strokes, and the 645Z can do that for images up to around 1.2m.

[eronald]

1000 per capture Edmund? Are you sure?   That means if one does 20 paintings ...a 20K day income?    Where is this museum? Can you send me an address of it?  

Usually a personal recommendation is necessary for these jobs, which of course means they have already been in touch with you  

Edmund

[Theodoros]

I recently bought a 645Z and do all my art repro with the FA 120mm Macro now, I used to stitch shots with a 5D before! But as far as printing on 44" canvas goes... it's not a particularly demanding medium in terms of resolution, nor does it have a particularly wide gamut, so a single shot is pretty much enough to reproduce a meter-wide painting in crisp detail, and going any bigger I would probably stitch two-three shots.

I think 80MP backs, scanning backs, and multi-shot systems may be nice for some real big artworks or getting that last 5% of image quality, but I tend to stop at the point where I can see bristle indents in the brush strokes, and the 645Z can do that for images up to around 1.2m.

Now wait a minute.... 5% it may be when one uses an 80mp back with colour interpolation instead of 645Z... Why not have a look at Torger's post on how an 80mp back compares with any multishot and "only" at 4x? Besides, getting an old Sinarback 54H or Imacon 528c will be considerably cheaper than using a 645Z ...no? ...and the difference is no where near 5%! ...it's far, ...far more distant than that... Why not try one before you judge what is "good enough"?  And more than that... what is not "particularly demanding nor has particularly wide Gamut"?  Have ever tried some Caravaggio or a Byzantine Icon that uses gold or silver?

[Theodoros]

Usually a personal recommendation is necessary for these jobs, which of course means they have already been in touch with you  

Edmund

I still don't see the relevance of the link Ed... I can assure you prices are no where near to 1000/per item... It's not a fee that one can complain for (far from that), It rather is one of the best (if not the best) paying sections in photography, but one must consider that skills are too time consuming to be achieved and the judges of the job are not the ones that can be fooled.... You see, with art reproduction, the recipient wants exactly that... a reproduction... Not just a file or a photograph of a painting! And of course the recipients are far from being ignorants on if the codes of the piece of art under examination are "transferred" or not!

[Kolor-Pikker]

Now wait a minute.... 5% it may be when one uses an 80mp back with colour interpolation instead of 645Z... Why not have a look at Torger's post on how an 80mp back compares with any multishot and "only" at 4x? Besides, getting an old Sinarback 54H or Imacon 528c will be considerably cheaper than using a 645Z ...no? ...and the difference is no where near 5%! ...it's far, ...far more distant than that... Why not try one before you judge what is "good enough"?

I've tried lots of systems before buying, and the 645Z is good enough for practical real-world reproduction printing, the prints I do can sell for thousands and that's enough of a standard for me.  Additionally, all other options would result in an application-specific camera that I wouldn't be able to use for anything other than shooting certain subjects in the studio, whereas instead I have camera that's as great in the studio as it is hand-held at ISO12,800, if I had bought two cameras for different subject matter, it would've ended up being more expensive. Maybe someone else may only need a camera that does only one thing well, and in that case perhaps the other options would be better.

And more than that... what is not "particularly demanding nor has particularly wide Gamut"?  Have ever tried some Caravaggio or a Byzantine Icon that uses gold or silver?

...Obviously I meant the canvas you print on, not the medium you photograph, I don't think it's possible to source a printer and inks that can reproduce gold or silver ink on wood, if even accurately. But yes, there are some artworks out there that the best cameras have trouble with.

[rubencarmona]

Actually to profile your camera to a reference works well enough for me when reproducing fine art. We also make pictures for an art gallery with a 5D Mark II and use a SpyderCheckr for color reproduction.

[eronald]

I still don't see the relevance of the link Ed... I can assure you prices are no where near to 1000/per item... It's not a fee that one can complain for (far from that), It rather is one of the best (if not the best) paying sections in photography, but one must consider that skills are too time consuming to be achieved and the judges of the job are not the ones that can be fooled.... You see, with art reproduction, the recipient wants exactly that... a reproduction... Not just a file or a photograph of a painting! And of course the recipients are far from being ignorants on if the codes of the piece of art under examination are "transferred" or not!

I must have included the link by mistake, while reading up on the trial of a photographer who *allegedly* extracted $500 million from his sponsor. I know the number seems strange, but my impression is that there is now a macroscopic world out there with giants who stand miles tall and count in large numbers, while we bacteria live in the economics of small numbers.

I would assume billing of around $1K per day here in northern Europe, for anyone with good connections to a major institution or a for a well-connected associate of an auction house. Of course someone working for  "normal" artists may earn 20% of that,  but can get part of his take in cash.

Edmund

[Theodoros]

I must have included the link by mistake, while reading up on the trial of a photographer who *allegedly* extracted $500 million from his sponsor. I know the number seems strange, but my impression is that there is now a macroscopic world out there with giants who stand miles tall and count in large numbers, while we bacteria live in the economics of small numbers.

I would assume billing of around $1K per day here in northern Europe, for anyone with good connections to a major institution or a for a well-connected associate of an auction house. Of course someone working for  "normal" artists may earn 20% of that,  but can get part of his take in cash.

Edmund

Yeap, that's more like it... although prices are always per piece photographed and the price varies with the size and if there is some complexity involved, usually the day income for a pro on the job is around 1500/per day of work (8 hours - the ones that can work more efficiently may sometimes exceed 2k/per day, or the opposite for the less capable ones). OTOH, "real connections" are more with private galleries or collectors those days and are based on fame, never the less, everybody is prepared to pay a little more (or less sometimes) and overcome the usual "connection" if quality offered by a new "contact" proves to be better... You see the better quality, directly affects the financial benefit of the recipient.
Usually, for major work with museums, there is an open competition announced and "standards" set for quality.... Then, there is an offering file applied and the best quality/price balance wins the competition... Where I live (the country that is blessed to be the undeniable world champion in civilisation treasures on the world), the directing committee for each of the thousands of museums and the preservation organisations that are responsible for ancient Byzantine monasteries and churches or other "open space" material, is aloud (under law) to only decide on giving a job directly to a photographer, if the total cost of the project comes down to under 20K... Anything above that, is an open competition...

[yaya]

Nobody claimed that a Bayer CFA image is 'better' than a micro-step (or a color-wheel with a monochrome sensor) capture.

Tri-chromatic (R/G/B) color capture is still a relatively rough approximation of the full visible spectrum reflectance ...
In that sense, even micro-step sensor capture is also 'interpolated' color, although more predictable than from a single Bayer CFA. It does require extremely constant continuous lighting, where temperature and long exposure time are enemies of delicate artwork.

+1

[ErikKaffehr]

Hi,

Regarding multishot and one shot, I would make the point that Bart van der Wolf has made some calculations and found that loss of resolution due to Bayer interpolation is around 10%. If we look at 4X multishot we can thus gain about 10% in resolution. So a 50 MP 4X multishot back would correspond to a 60 MP one shot.

With 16X multishot the back will utilise the interpixel gap, and can reach 4x times the resolution, so it would correspond to 200 MP. With CCD sensors there is quite obviously an interpixel gap, else microlenses would give little sensivity gain.

Multishot obviously eliminates colour moiré. On the other hand, aliasing and moiré will not arise as long as the sensor outresolves the subject or the lens.

Resolution can be increased by stitching turning the camera vertical and stitching horizontally. That is another option.

Best regards
Erik

[Bart_van_der_Wolf]

Regarding multishot and one shot, I would make the point that Bart van der Wolf has made some calculations and found that loss of resolution due to Bayer interpolation is around 10%.

Hi Erik,

For luminance resolution that's correct, although for an unlikely combination of particular colors (exact same luminance and perfect chrominance matching to the R/B filters transmission on the CFA) the resolution could be 50%. The demosaicing test can be found here.

This is for the loss due to demosaicing only, the interaction between the optical system and the possibilities of the demosaicing algorithms will determine the actual loss of resolution after MTFs have been combined into a system MTF.
 
Stitching at a higher capture magnification (e.g. longer focal length) and down-sampling that, will solve most of the potential resolution issues and also reduce aliasing because original image detail is sampled at a larger magnification relative to the sensor's sampling pitch.

Cheers,
Bart

P.S. I've attached a sample crop from a 235 Megapixel stitched reproduction, the original painting's size of the crop would be some 73x73mm. I also added a 50% down-sampled crop from the same location (146x146mm) of the painting, which just loses more real luminance resolution than it gains color precision. Accurate color is a matter of profiling, which is a whole other can of worms.

[Theodoros]

 In Art reproduction (not pictures of art) there is three factors that must be mastered...
 
ONE is setting up the subject and lighting it correctly.
TWO is having all the detail of it (for 1:1 print).
THREE is having all the tone and colour accuracy of it.

ONE differs between subjects and structure of the subject. TWO depends on the size and detail on the subject. THREE requires having absolute profiles... As far as "absolute" profiles are concerned, one can't have the required accuracy if colour interpolation is involved and "automatic" profiling "general" methods are not sufficient for having the term "reproduction" satisfied... Thus, the use of "true colour" is a necessity (to avoid an interpolation algorithm to interfere) and there is special knowledge required, where one has to "build" his own profiles for a particular project... That said, there are cases where one has to "group" the project in order to keep lighting and profile changes to a minimum.

Detail is of course as important as the other factors (but not more than the others), but one has to remember that resolution is only one part of detail (though important), the other parts of detail (tonality - DR), are equally important.

As far as resolution is concerned, the use of tri-colour capture clearly benefits than having a same size, same pixel density sensor with colour interpolation involved (since human eye "understands" as resolution the colour difference between pixels), but the advantage can't be measured in value... it can range from noticeable to huge depending on the complexity of the subject.

Of course tri-colour capture is not perfect... but it's the best we have available to work with and (most importantly) it has no disadvantage (only advantages with all factors involved) than the use of "normal" sensors. The longer exposure times and extensive lighting on the subject are factors that are "part of the job" for one to learn on how to control as to avoid harming the subject by emitting extensive IR on it.

[eronald]

Bart,

 As a consumer it is easy to experimentally see the difference between a Bayer file and a non-Bayer file by just comparing a $400 Sigma file to an SLR.

 I don't know why nobody makes a modern filter wheel camera for repro - with a mono version of a modern high-ISO CMOS sensor you'd just need to bring up a stabilised illuminant source for 5 seconds or so to get a good multispectral capture. Most display environments subject paintings to visible light

edmund

Hi Erik,

For luminance resolution that's correct, although for an unlikely combination of particular colors (exact same luminance and perfect chrominance matching to the R/B filters transmission on the CFA) the resolution could be 50%. The demosaicing test can be found here.

This is for the loss due to demosaicing only, the interaction between the optical system and the possibilities of the demosaicing algorithms will determine the actual loss of resolution after MTFs have been combined into a system MTF.
 
Stitching at a higher capture magnification (e.g. longer focal length) and down-sampling that, will solve most of the potential resolution issues and also reduce aliasing because original image detail is sampled at a larger magnification relative to the sensor's sampling pitch.

Cheers,
Bart

P.S. I've attached a sample crop from a 235 Megapixel stitched reproduction, the original painting's size of the crop would be some 73x73mm. I also added a 50% down-sampled crop from the same location (146x146mm) of the painting, which just loses more real luminance resolution than it gains color precision. Accurate color is a matter of profiling, which is a whole other can of worms.