Scanning with digital camera and stitching

JWMster · May 6, 2020

FWIW, Fuji cameras have mostly 24MP and take shots I'd argue as good as the Sony AR7 series of cameras. The extra megapixels in the Sony simply provide more opportunity for dust to land on the sensor. Just kidding. IMHO, Sony's chief advantage lay in their souped up stabilization and (at the time I used them) a wider set of lenses with which they could work, but otherwise, I was happy with both. Zeiss and otheers were making lenses for Sony and not Fuji so I went in that direction (for a while). Lens lust led to a trade out (Fuji) and in (Fuji and then Sony), but also to the conclusion that 24mp was all I needed. Consolidated my shooting 35mm and digital with a Nikon family, and held the budget to a D750. Mirrorless is easier to scan with in setting focus with "12X zoom" for that purpose, but DSLR works fine. Point is simply that with either a shift to MF or LF, the worry about grain clusters recedes and the emphasis shifts to "how fast can I get this scanning thing over with?" - and that's where I'm at.

Adrian Bacon · May 6, 2020

wahiba said:
I scan and stitch 120 images using the 35mm facility of my Epson V300 and PE7. I also make up panoramas from digital and analogue and unless I have left no overlap, or far too little for the software to work on I have yet to find the software join.

I don’t take issue with scanning and stitching per se. It has its uses, and can provide pretty good results. I personally use it when scanning panorama negatives and the like. It’s very useful in that type of use case, especially if you can easily maintain the overall total system performance, in that scenario, it’s a total no brainer if you end up with a negative that has a different aspect ratio (like a panorama) to just keep everything the same and capture the image over a number of shots and stitch it together. That’s very useful.

what I take issue with mostly is doing scanning and stitching in an effort to overcome perceived issues and limitations with a given system or technology, especially when an easier path to higher quality exists (like simply getting a higher resolution sensor, or using a larger negative, or getting a better lens, etc). People are absolutely free to do what they want (and they should), but they should really make the distinction that they’re doing it because they want to investigate doing it that way and document the results, not hail it as THE solution to some problem. But, that is just me.

Adrian Bacon · May 6, 2020

JWMster said:
Point is simply that with either a shift to MF or LF, the worry about grain clusters recedes and the emphasis shifts to "how fast can I get this scanning thing over with?" - and that's where I'm at.

that’s where I’ve been at for quite a long while as well. 24-32MP in a single capture with a really good lens and light source gives very fast results that are very serviceable for the vast majority of uses. I’ve actually run into a different problem. I have a number of clients that love the quality of my scans, but would rather just have jpegs simply because they are smaller files. Some have even asked for jpegs that they can fit on a thumb drive so they can put the pictures on as a slideshow on their TV. Assuming they have a 4K TV, at a 32MP scan, I’m basically oversampling quite a lot with my scans, and 35mm film at 4K resolution looks shockingly good on a big 60+ inch 4K TV. Makes you wonder how much resolution you really need to have, even for big prints.

alanrockwood · May 6, 2020

This has been an interesting discussion, though it seems to have gotten off-track a little. This is not necessarily bad, but it somewhat misses the point of my original posts. My main point in posting was to show how it is possible to get extremely high dpi (and presumably very high resolution as well) using very simple and inexpensive equipment.

In this case I have spent well under $100 in equipment and parts, partly because I used things I already had, but even if I would have bought all of the parts it would not have cost me very much, far less, for example, than something like a FlexTight or drum scanner or even a Nikon scanner, and the Nikon scanner would not be able to touch what this approach is capable of. (This is yet to be proven, but I am confident that it is true.) Of course, a Nikon or FlexTight is a lot more convenient, if one has the money to buy one.

It is also only a tiny fraction of the price of a dslr with a high megapixel rating.

Adrian Bacon · May 6, 2020

alanrockwood said:
This has been an interesting discussion, though it seems to have gotten off-track a little. This is not necessarily bad, but it somewhat misses the point of my original posts. My main point in posting was to show how it is possible to get extremely high dpi (and presumably very high resolution as well) using very simple and inexpensive equipment.

In this case I have spent well under $100 in equipment and parts, partly because I used things I already had, but even if I would have bought all of the parts it would not have cost me very much, far less, for example, than something like a FlexTight or drum scanner or even a Nikon scanner, and the Nikon scanner would not be able to touch what this approach is capable of. (This is yet to be proven, but I am confident that it is true.) Of course, a Nikon or FlexTight is a lot more convenient, if one has the money to buy one.

It is also only a tiny fraction of the price of a dslr with a high megapixel rating.

no arguments from me. How much time has it cost you, and what is your time worth?

Les Sarile · May 6, 2020

There are reasons for Coolscan's price.
With higher resolution DSLR copying the more porminent the dust and scratches will show. A huge advantange for Coolscan's ICE.
DSLR copying of color negatives require a conversion process. I have yet to see any that is remotely as quick or as good as Coolscan+Nikonscan. A Coolscan 5000 scan takes about 50 seconds per frame with ICE- about 30 seconds without.
Coolscan 5000 has motorized accessories for batch scanning of many slides or whole rolls of film.
As I have already shown, the Coolscan's 4000dpi is more closely a match for at least a 36MP DSLR.
As much as I appreciate the high resolving power of the Coolscan's 4000dpi - I have compared 20" X 30" professionally done optical poster prints from 35mm film, I value the color/contrast fidelity more.

Helge · May 7, 2020

Adrian Bacon said:
exactly. The point I was trying to make is that the physical film itself is one piece of a total system. All the little bits and pieces affect total system resolution and it’s a lot better to focus on improving the total system or start with a reasonably good system and have serviceable images for the vast majority of uses. For 35mm film, a 24MP bayer sensor coupled with a really good macro lens and a solid scanning set up and light source is pretty difficult to top without significantly increasing the amount of time and complexity required to get there, and I would argue that the improvement wouldn’t really be significantly better for why you would go for all that extra resolution, which is print super huge at super high resolution. Why would you do that with 35mm? You can get much better performance going to a larger negative size.

Why did 135 film become the dominant in the first place?
Some answers to the question is flexibility, speed (not having to reload every 8, 12, 16 or 24 frames), greater selection of useful, practical and affordable lenses.
For example the Hasselblad equivalent of a 200mm f4 is huge, a stop slower (hard to focus on ground glass) and costs ten times as much.
And crucially: It desperately needs a tripod or other support to use the resolution of the format, due to the focal length of 350 and weight of the whole setup. Schwarzenegger in his prime wouldn’t be able to hold it still.
The story is much the same with wide lenses.
Large, expensive, slow and fickle with how they can be used.

If it is possible to have very good and usable results including the above mentioned advantages, then why not use 135 film?
Even if it means slightly more effort has to be spent in the darkroom or doing scanning. A situation and setting where time, space an weight is not at such premium. And expense can be deferred or upgraded/franchised out as needed (better enlarger lenses, better scanner etc.)

Helge · May 7, 2020

Adrian Bacon said:
I’d love to see some actual evidence of this on film scanned with a bayer sensor. Both the raw scan before processing to get a color positive, and the processed color positive after.

the color filters are not unique to cameras. Scanners also have color filters that do the same thing.

Bayer sensors filters in cameras are often fainter in colour, to make demosaicing and interpolation easier/possible, at the expense of colour depth and resolution.
The very idea of the Bayer filter is already a step in that direction.
Simply have a look at the spectral response curves of film versus the filters used in the sensor.
Or for a more empirical example put a piece of slide film of a colorful scene on a light table.
Some of the difference will be due to the higher contrast than just about any other display technology, but the saturation and depth of colours is simply outstanding.

Helge · May 7, 2020

Les Sarile said:
There are reasons for Coolscan's price.
With higher resolution DSLR copying the more porminent the dust and scratches will show. A huge advantange for Coolscan's ICE.
DSLR copying of color negatives require a conversion process. I have yet to see any that is remotely as quick or as good as Coolscan+Nikonscan. A Coolscan 5000 scan takes about 50 seconds per frame with ICE- about 30 seconds without.
Coolscan 5000 has motorized accessories for batch scanning of many slides or whole rolls of film.
As I have already shown, the Coolscan's 4000dpi is more closely a match for at least a 36MP DSLR.
As much as I appreciate the high resolving power of the Coolscan's 4000dpi - I have compared 20" X 30" professionally done optical poster prints from 35mm film, I value the color/contrast fidelity more.

ICE is not free. It comes at the cost of slight but noticeable blurring.
The need for dust removal is largely an effect of using these kinds of holders, that attract dust like crazy.
It was never that much of an issue with enlarging and the little there was, could be taken care of with a blower.
Should it be pathological you can rewash the film.
ICE would in theory be fully possible with a DSLR, since CMOS sensors are also sensitive to IR.
The taking process is not what is holding DSLR scanning back.
That process takes from a second for a single shot, to a minute for the deluxe version with multiple shots, and at different apertures.

The difference is in the convenience of the software.

Negative Lab Pro is taking a stab at that, but is hindered by forcing prospects to buy into the monthly extortion from Adobe.
Stand alone please!

Les Sarile · May 7, 2020

Helge said:
ICE is not free. It comes at the cost of slight but noticeable blurring.
The need for dust removal is largely an effect of using these kinds of holders, that attract dust like crazy.
It was never that much of an issue with enlarging and the little there was, could be taken care of with a blower.
Should it be pathological you can rewash the film.
ICE would in theory be fully possible with a DSLR, since CMOS sensors are also sensitive to IR.

I am familiar with the limitations - as well as benefits, of it. The Coolscan+Nikonscan ICE is nothing short of magical and a compromise I fully accept considering the alternatives. I don't miss spotting dark room prints although I still have to spot my true b&w scans . . .

On Kodachrome . . .

On Kodak 160VC. . .

I couldn't spot these in PS anywhere near the the time it takes the Coolscan - about 50 seconds. Maybe I can come close to the quality of the removal if I worked on it for . . . days?

Nikon has incorporated color negative conversion into it's D850 so that would be nice if they can incorporate iICE into DSLR scanning.

138S · May 7, 2020

alanrockwood said:
OK, it's not a new idea, but I have been putting together a simple system to scan film using a microscope objective to capture small pieces of the film image and then stitching the results together.

I've also tried that, but you can reach total performance in a more productive way.

You can get 10,000 dpi optically effective by shoting only a 3x3 mosaic in a 35mm frame, use a cheap/excellent EL Nikkor 50mm enlarger lens reversed, in that way you have a better DOF. If not you may have practical problem to focus well.

Practical concerns are DOF and field curvature !!!

Get a USAF 1951 glass slide to measure well what performance you obtain.

JWMster · May 7, 2020

I have a Nikon Coolscan LS8000 and find the speed... or more correctly - the LACK of speed means that time I could use in editing the shots, adjusting the contrast, color temp, etc. is lost in the grinding of the motorworks. Scanning 120 film is about 20 minutes or so for 3 frames for prescan and scan. Ouch. The scanning of 300 plus MF shots from a trip to France? The scan that broke the camel's back. Yes Nikon Coolscans are good, but are they really better? Kenneth Lee Graham has some useful tidbits here ( http://www.kennethleegallery.com/html/tech/index.php ) but I'd add that the larger the format, the more it seems as though your attention to the details of keeping the film and film environment clean rise and the relatively less significant scaling of the dust-to-image size becomes. My 20-to-30-year-old (plus) scanner made the trip to the new house, but remains unpacked, and I'm hoping it will stay in its comfort zone. Negative Lab Pro plus Negative Solutions gear is a "better" answer to today's needs the way that 35mm fit the on-the-go needs of the 1960's: Not initially better per se, but enabling a new portability that paid dividends down the road.

Let's not get lost in the weeds here: Nikon and other scanners are/were fine and may serve today with some catch-up maintenance (mine took a bunch of $'s to put back into shape and keep there), but for the vast majority of new-to-filmers these are likely to prove of questionable ECONOMIC merit for the years ahead.

MattKing · May 7, 2020

Helge said:
The need for dust removal is largely an effect of using these kinds of holders, that attract dust like crazy.
It was never that much of an issue with enlarging and the little there was, could be taken care of with a blower.

It isn't just the holders, it also is a characteristic of the optical path and light source.
It reminds me of working with a condenser enlarger and a small bulb light source. I expect that if one used a point source enlarger, it would be even more similar.

Les Sarile · May 7, 2020

JWMster said:
I have a Nikon Coolscan LS8000 and find the speed... or more correctly - the LACK of speed means that time I could use in editing the shots, adjusting the contrast, color temp, etc. is lost in the grinding of the motorworks. Scanning 120 film is about 20 minutes or so for 3 frames for prescan and scan. Ouch. The scanning of 300 plus MF shots from a trip to France? The scan that broke the camel's back. Yes Nikon Coolscans are good, but are they really better? Kenneth Lee Graham has some useful tidbits here ( http://www.kennethleegallery.com/html/tech/index.php ) but I'd add that the larger the format, the more it seems as though your attention to the details of keeping the film and film environment clean rise and the relatively less significant scaling of the dust-to-image size becomes. My 20-to-30-year-old (plus) scanner made the trip to the new house, but remains unpacked, and I'm hoping it will stay in its comfort zone. Negative Lab Pro plus Negative Solutions gear is a "better" answer to today's needs the way that 35mm fit the on-the-go needs of the 1960's: Not initially better per se, but enabling a new portability that paid dividends down the road.

Let's not get lost in the weeds here: Nikon and other scanners are/were fine and may serve today with some catch-up maintenance (mine took a bunch of $'s to put back into shape and keep there), but for the vast majority of new-to-filmers these are likely to prove of questionable ECONOMIC merit for the years ahead.

I have the Coolscans V, 5000 and 9000 and they all greatly improved on the scanning speeds of the previous version. I wish it could be faster. Certainly the scan from a DSLR is faster - I have autobellows from 35mm, but to get anywhere near the results I get from Coolsca+Nikonscan takes far more time. Also, there were a lot of online posts complaining about scan lines and bad focus on the previous generation but not on these. So at least they are definitely better in those regards.

With regards to people new to film, the advantages of the Coolscan+Nikonscan are even more important. Most people new to film simply get cheapest develop and scan leading to complaints of poor color or blown out highlights. You know how hard it is to blow out the highlights on most films that are not slide?

Here's an example of the same frame of Kodak Gold 100 scanned on a Noritsu minilab vs fully automatic scan on Coolscan+Nikonscan with all color settings off with no pre or post adjustment.

Same frame of Kodak 160VC at a Fuji Frontier minilab vs Coolscan+Nikonscan with all color settings off with no pre or post adjustment.

They wouldn't even know that film has dust and scratches if they use the Coolscan+Nikonscan . . .

Adrian Bacon · May 8, 2020

Helge said:
Bayer sensors filters in cameras are often fainter in colour, to make demosaicing and interpolation easier/possible, at the expense of colour depth and resolution.

can you point me to some documentation on the web somewhere that backs this up? I’ve long studied bayer sensors and would love to read any research done in regards to its color/spatial performance.

that being said, how exactly does a fainter color filter make it easier to demosaic or interpolate? I’m asking because I’ve written a fair amount of code that does exactly that and I’d imagine that the color filters do vary from camera model to camera model, and for some reason the algorithm to demosaic the sensor is exactly the same between camera models.

what it does effect is the color matrix used to conform the raw colors to a color space, however once you’ve characterized that and have a color matrix, the color performance tends to be very high. You also have to do the same thing when you have a full RGB per pixel like in a scanner. The only difference is no demosaic step.

alanrockwood · May 8, 2020

Adrian Bacon said:
can you point me to some documentation on the web somewhere that backs this up? I’ve long studied bayer sensors and would love to read any research done in regards to its color/spatial performance.

that being said, how exactly does a fainter color filter make it easier to demosaic or interpolate? I’m asking because I’ve written a fair amount of code that does exactly that and I’d imagine that the color filters do vary from camera model to camera model, and for some reason the algorithm to demosaic the sensor is exactly the same between camera models.

what it does effect is the color matrix used to conform the raw colors to a color space, however once you’ve characterized that and have a color matrix, the color performance tends to be very high. You also have to do the same thing when you have a full RGB per pixel like in a scanner. The only difference is no demosaic step.

This link may contain some of the information you are interested in . It is a database of spectral sensitivity curves measured for various cameras.

https://nae-lab.org/~rei/research/cs/zhao/database.html

Here's a link to their result for the Canon XTi,

https://nae-lab.org/~rei/research/cs/zhao/files/canon_400d.jpg

and here's a link to their results for the Sony dxc 9000.

https://nae-lab.org/~rei/research/cs/zhao/files/sony_dxc_9000.jpg

Helge · May 8, 2020

Adrian Bacon said:
can you point me to some documentation on the web somewhere that backs this up? I’ve long studied bayer sensors and would love to read any research done in regards to its color/spatial performance.

that being said, how exactly does a fainter color filter make it easier to demosaic or interpolate? I’m asking because I’ve written a fair amount of code that does exactly that and I’d imagine that the color filters do vary from camera model to camera model, and for some reason the algorithm to demosaic the sensor is exactly the same between camera models.

what it does effect is the color matrix used to conform the raw colors to a color space, however once you’ve characterized that and have a color matrix, the color performance tends to be very high. You also have to do the same thing when you have a full RGB per pixel like in a scanner. The only difference is no demosaic step.

I'd be hard pressed to do it off the cuff, without sinking an hour or two into digging the sources out.
Alans links looks good though.
Also do a search on David Mullen the cinematographer. He shoots both film with zest and is a RED ambassador.. Especially noteworthy is his postings on the ASC site. He has quite a good technical grip on what goes on in the sensor and immediately after when the image is read off the sensor and importantly in this context the advantages of film with colour registration.

The idea of having a non fully saturated RGB filter array, is to make it easier to extrapolate luminance information, without guessing so much.
The bet is that you can still make a reasonable guess at colours with better demosaicing, and people will prefer the apparent higher resolution to better colour precision when they pixel peep.

As mentioned, somewhat the same idea as having more green sensor sites to up the spatial resolution, just taken a step further.
Some manufacturers have even experimented with CMY instead of RGB for the same reason, with little success though.

Colour resolution is kind of more ephemeral than luminance, especially to explain with words. Show some clear visual examples and it becomes clearer. And the math behind it, is more or less bullet proof.

Lachlan Young · May 8, 2020

Adrian Bacon said:
can you point me to some documentation on the web somewhere that backs this up? I’ve long studied bayer sensors and would love to read any research done in regards to its color/spatial performance.

that being said, how exactly does a fainter color filter make it easier to demosaic or interpolate? I’m asking because I’ve written a fair amount of code that does exactly that and I’d imagine that the color filters do vary from camera model to camera model, and for some reason the algorithm to demosaic the sensor is exactly the same between camera models.

what it does effect is the color matrix used to conform the raw colors to a color space, however once you’ve characterized that and have a color matrix, the color performance tends to be very high. You also have to do the same thing when you have a full RGB per pixel like in a scanner. The only difference is no demosaic step.

The colour performance difference between the trichromatic Phase One back and the regular filter array back is pretty obvious - as is the serious speed loss incurred - in the order of 2 2/3 stops for the trichromatic back over the achromatic/ unfiltered back vis-a-vis the approx 1 stop loss most Bayer array filters incur. I suspect that the filters used on the trichromatic are closer in transmission peaks to a standard separation filter set. I suspect this also why 3xCCD/ PMT scanners etc were seen as giving better colour too, because they didn't have to compromise in their transmittance peaks (and try to correct in software) to squeeze out a little more speed to comply with market demands for acceptable baseline ISO as cameras did. Otherwise, Nikon/ Canon etc would have been stuck with base ISO in the 25-64 range even now.

Adrian Bacon · May 8, 2020

Helge said:
I'd be hard pressed to do it off the cuff, without sinking an hour or two into digging the sources out.
Alans links looks good though.
Also do a search on David Mullen the cinematographer. He shoots both film with zest and is a RED ambassador.. Especially noteworthy is his postings on the ASC site. He has quite a good technical grip on what goes on in the sensor and immediately after when the image is read off the sensor and importantly in this context the advantages of film with colour registration.

The idea of having a non fully saturated RGB filter array, is to make it easier to extrapolate luminance information, without guessing so much.
The bet is that you can still make a reasonable guess at colours with better demosaicing, and people will prefer the apparent higher resolution to better colour precision when they pixel peep.

As mentioned, somewhat the same idea as having more green sensor sites to up the spatial resolution, just taken a step further.
Some manufacturers have even experimented with CMY instead of RGB for the same reason, with little success though.

Colour resolution is kind of more ephemeral than luminance, especially to explain with words. Show some clear visual examples and it becomes clearer. And the math behind it, is more or less bullet proof.

OK, well, if you find the time and want to do it, I think it’d make for an interesting discussion.

I read quite a bit of what Mullen posts. He dispenses solid info.

in terms of the whole luminance thing you’re referring to I’m scratching my head a bit... Luminance values can’t be had until you have a full rgb value at each pixel site, which doesn’t happen until after you’ve demosaiced the filter. Many people say the demosaic is a bunch of guessing, and technically, they’d be right, however, it’s an extremely good guess, as you are in fact sampling something at every sensel site and depending on the algorithm you’ve chosen to demosaic with, you’re not just naively filling in the holes.

the demosaic code I’ve written is based off of the AHD algorithm, but I’ve optimized it to be a lot more intelligent with respects to preserving edges (significantly beefed up the edge detection, much, much larger neighborhood size, interpolate along detected edges, not across them, etc) and I very much prefer using uniform centripetal catmull rom for interpolation over other interpolation routines. When interpolating along edges, it just looks a lot better. Other interpolation algorithms are better suited for general image resizing (I’ve written and tested a lot of them), but I’ve gotten best results with catmull-rom when demosaicing a bayer array.

once I have a full rgb value at each pixel, luminance is pretty straightforward, as that algorithm is standardized and published.

Helge · May 9, 2020

Of course demosaicing is a good guess, and overall an impressive feat, considering what the initial dataset looks like to the human eye.
But there is only so many ways you can improve it. Even new deep learning networks approaches, sees only very marginal improvements.
The main problem persists as smudging or waxiness of complex real world structures and textures. IE not USAF 1951 or picket fences.
And, a subtle but overall noticeable (if only subconsciously), different treatment of high contrast, clear geometric edges. Sometimes appearing as almost unsharp masking.

The trouble is of course, that people like "clean" and they love "sharp".
They've been taught to by the photo industry, and billions of examples, and the human visual system gobbles clean and sharp up as "free processing".
It's the equivalent of too much sugar and salt, on small samplers of food.
It's nice in small doses, but becomes cloying and too much all the time and in meal sized portions.
Of course, looking at images shouldn't be "eating your vegetables". So sure, sharpening and cleaning has its place, but everyone needs a healthy varied diet.
You don't see any problem with too clean and too sharp until you acclimatize, and look at how good big film prints used to look only a decade or two ago.

All very academic since your scans are probably terrific.
Again, what I'm talking about is that one in a hundred or a thousand frames, that you want to print at some size and present on your wall for years or use in a gallery.

Regarding luminance and bayer filtering:
I'm having trouble seeing what's puzzling you. You basically describe yourself what I'm trying to convey: "you are in fact sampling something at every sensel site and depending on the algorithm you’ve chosen to demosaic with, you’re not just naively filling in the holes".
Less aggressively saturated colour filters will make continuity and rapid changes in luminance dominated, monochrome and even very complexly detailed multi chromatic structures easier to "glean".

Stitching at higher macro settings will not only give you the needed resolution to get over Nyquist and grain aliasing, it will also also allow somewhat higher colour precision.

Adrian Bacon · May 9, 2020

alanrockwood said:
This link may contain some of the information you are interested in . It is a database of spectral sensitivity curves measured for various cameras.

https://nae-lab.org/~rei/research/cs/zhao/database.html

Here's a link to their result for the Canon XTi,

https://nae-lab.org/~rei/research/cs/zhao/files/canon_400d.jpg

and here's a link to their results for the Sony dxc 9000.

https://nae-lab.org/~rei/research/cs/zhao/files/sony_dxc_9000.jpg

thanks for the links, I’m still reading through it, but have some initial concerns... they don’t appear to be looking directly at raw sample data. I say this because they make no mention of what raw multipliers they’re using for each color channel, and that will effect the outcome a lot. Every digital camera has a native white balance (along the amber/blue axis) where the multipliers between the red and green channels will be the same. The same for the green/magenta axis. If you don’t use the same multipliers between the color channels, you really significantly distort what the color response would be.

out of the box, if you use 1.0 for the multipliers for all three channels, you’ll quickly discover that the green channel almost universally is most sensitive and has the highest response (which is not reflected by the data you are referencing) relative to the other two channels. You’ll also discover that there is a spot on the WB Kelvin scale that produces the same output between the red and blue channels, and that manufacturers tune the color gels that they put in the bayer array to better suit how the camera is intended to be used. For example, the Canon 80D (a camera I’ve studied very intensively as it was my main scanning camera from its release to just a few months ago) is intended to be a general purpose camera. It’s native white balance is 5200K. A good all around general purpose WB. If you shoot under light that is 5200K, it has excellent color performance and dynamic range. Conversely, the Canon EOS RP has a native white balance of 4500K. Something a little better tuned to shooting indoors under artificial light. A camera intended for studio use with flash will have a native white balance closer 6500-7000K, and one intended to shoot video under tungsten light will have one closer to 3200K.

my point is, the linked info is interesting, and I’m not saying it’s invalid for their stated use case, I’m just a little concerned with how they’re going about gathering the data. Btw, I have a digital rebel xti, though haven’t really bothered to look closely at its WB response.

Grim Tuesday · May 9, 2020

Guys, I have a V700, a Nikon 9000ED and a DSLR scanning setup that supports single shot and stitched capture I was quite happy with before I started buying scanners. I've been meaning to scan the same negative with all three for a while now but haven't gotten around to it. Tomorrow seems as good a day as any. Is there an particular procedures I should follow in my test that you guys would recommend so that it will, if not answering this question once and for all, add a good data point to the discussion?

Helge · May 9, 2020

Grim Tuesday said:
Guys, I have a V700, a Nikon 9000ED and a DSLR scanning setup that supports single shot and stitched capture I was quite happy with before I started buying scanners. I've been meaning to scan the same negative with all three for a while now but haven't gotten around to it. Tomorrow seems as good a day as any. Is there an particular procedures I should follow in my test that you guys would recommend so that it will, if not answering this question once and for all, add a good data point to the discussion?

That's a very nice offer!

I'd say scan 135/35mm to put a point on it.
The film is the same per mm, but the demand on the scanner optics is not.

Use flash/speedlight backlight if possible, with some blue Wratten/RA4 compensation filters over the light aperture mask, if you have them.
This makes life easier when removing the orange mask.

Sandwich the film between glass, the thinner the better on top. Or optical grade film on top.
Newton rings be damned, if you don't have any suitable scanning fluid. We can abstract from them.
It's more important to keep the film completely flat.

Tether the camera to your computer, so you can use a decent sized monitor to get focus bang on.

You can use HDR techniques to get all the dynamics out. Take many exposures of the same part at different times or apertures to get all the shadow detail.
With negative film, the problem is not as with slide, punching through/DMax.
It's not having the shadow detail bloomed out and veiled over.

But of course the above is for the ultimate DSLR scan. It could take some time to get through the whole procedure.

Helge · May 9, 2020

Adrian Bacon said:
thanks for the links, I’m still reading through it, but have some initial concerns... they don’t appear to be looking directly at raw sample data. I say this because they make no mention of what raw multipliers they’re using for each color channel, and that will effect the outcome a lot. Every digital camera has a native white balance (along the amber/blue axis) where the multipliers between the red and green channels will be the same. The same for the green/magenta axis. If you don’t use the same multipliers between the color channels, you really significantly distort what the color response would be.

out of the box, if you use 1.0 for the multipliers for all three channels, you’ll quickly discover that the green channel almost universally is most sensitive and has the highest response (which is not reflected by the data you are referencing) relative to the other two channels. You’ll also discover that there is a spot on the WB Kelvin scale that produces the same output between the red and blue channels, and that manufacturers tune the color gels that they put in the bayer array to better suit how the camera is intended to be used. For example, the Canon 80D (a camera I’ve studied very intensively as it was my main scanning camera from its release to just a few months ago) is intended to be a general purpose camera. It’s native white balance is 5200K. A good all around general purpose WB. If you shoot under light that is 5200K, it has excellent color performance and dynamic range. Conversely, the Canon EOS RP has a native white balance of 4500K. Something a little better tuned to shooting indoors under artificial light. A camera intended for studio use with flash will have a native white balance closer 6500-7000K, and one intended to shoot video under tungsten light will have one closer to 3200K.

my point is, the linked info is interesting, and I’m not saying it’s invalid for their stated use case, I’m just a little concerned with how they’re going about gathering the data. Btw, I have a digital rebel xti, though haven’t really bothered to look closely at its WB response.

Well, as you probably noticed already, something interesting though not surprising, can be observed if you compare between the response curves for the 3 CCD cameras and the Bayered CCD/CMOS sensors.
I can almost guarantee you that things have not improved since. Probably on the contrary.
Colour resolution is not as heavily advertised, as resolution, low light performance and various tricks to get better "dynamics".
Colour "correctness" is often substituted, which of course is mostly BS, as I'm sure you'll agree.

Then compare the curves with the colour response curves of Kodak and Fuji film.

alanrockwood · May 9, 2020

Adrian Bacon said:
thanks for the links, I’m still reading through it, but have some initial concerns... they don’t appear to be looking directly at raw sample data. I say this because they make no mention of what raw multipliers they’re using for each color channel, and that will effect the outcome a lot. Every digital camera has a native white balance (along the amber/blue axis) where the multipliers between the red and green channels will be the same. The same for the green/magenta axis. If you don’t use the same multipliers between the color channels, you really significantly distort what the color response would be.

out of the box, if you use 1.0 for the multipliers for all three channels, you’ll quickly discover that the green channel almost universally is most sensitive and has the highest response (which is not reflected by the data you are referencing) relative to the other two channels. You’ll also discover that there is a spot on the WB Kelvin scale that produces the same output between the red and blue channels, and that manufacturers tune the color gels that they put in the bayer array to better suit how the camera is intended to be used. For example, the Canon 80D (a camera I’ve studied very intensively as it was my main scanning camera from its release to just a few months ago) is intended to be a general purpose camera. It’s native white balance is 5200K. A good all around general purpose WB. If you shoot under light that is 5200K, it has excellent color performance and dynamic range. Conversely, the Canon EOS RP has a native white balance of 4500K. Something a little better tuned to shooting indoors under artificial light. A camera intended for studio use with flash will have a native white balance closer 6500-7000K, and one intended to shoot video under tungsten light will have one closer to 3200K.

my point is, the linked info is interesting, and I’m not saying it’s invalid for their stated use case, I’m just a little concerned with how they’re going about gathering the data. Btw, I have a digital rebel xti, though haven’t really bothered to look closely at its WB response.

As I look at the data on their web page and also reading the paper they reference ("https://nae-lab.org/~rei/research/cs/zhao/files/MIRU09.pdf") As I look closer at their web page I believe you are right. I don't think they are reporting the data directly. I don't know about the curve normalization issue you raise when you mentioned multipliers, but it looks like they might be fitting their data to basis sets derived from singular value decompositions. There is not enough information to state this for sure, but if it is true then their data might have some errors relative to the data acquired directly from the cameras. For example, their reported curves are might be a little smoother than the underlying data and may miss some of the small features in the data, e.g. they might not correctly represent the data in the tails of the curves or some small bumps and wiggles in the curves.

Scanning with digital camera and stitching

Moderator

Privacy & Transparency

Privacy & Transparency