Digitizing Negatives

Hello, all, hope this is the right place to post my question. I've been shooting digital for 20+ years but have recently acquired a few film cameras and now need to digitize my negatives. I've decided on using my Canon 5DsR (50mpx) camera and my  Canon 100mm f/2.8 macro (the non-L) version. My question concerns the method of digitizing the negatives.

I have found on Amazon a Lens Tube mechanism that attaches to the lens and should work well with a light panel that I have. Alternatively, there is also the Copy Stand method which also seems to work well with a light panel. Click the two links to see the Amazon entries.

So, as between the two is there a consensus of the better product?

Finally, I have a Manfrotto tripod w a head that will tilt in such a way that I could conceivably use that instead.

Any thoughts and insights will be greatly appreciated.

AceVentura0212 wrote:

Hello, all, hope this is the right place to post my question. I've been shooting digital for 20+ years but have recently acquired a few film cameras and now need to digitize my negatives. I've decided on using my Canon 5DsR (50mpx) camera and my Canon 100mm f/2.8 macro (the non-L) version. My question concerns the method of digitizing the negatives.

I have found on Amazon a Lens Tube mechanism that attaches to the lens and should work well with a light panel that I have.

This is most similar to what I use, which is the 3D-printed DupliHood typically on my Macro Rokkor 50mm f/3.5. Consider using your camera's HDR mode (if it has one). I'm not sure how the unit you link to would work with a 100mm macro -- I don't see much adjustment of the adapter's length.

Alternatively, there is also the Copy Stand method which also seems to work well with a light panel. Click the two links to see the Amazon entries.

Can work too. Should be more versatile, but much harder to get everything aligned.

So, as between the two is there a consensus of the better product?

In either case, don't have the light panel too close behind the negative, because that will help evenness of the illumination. Also be careful about shutter speed, because LED panels use PWM that can cause banding at some speeds.

Finally, I have a Manfrotto tripod w a head that will tilt in such a way that I could conceivably use that instead.

The really slow to align option...

Any thoughts and insights will be greatly appreciated.

I actually own a Nikon CoolScan 9000, which is one of the best dedicated scanners ever made, but easily get better quality using a Sony A7 with the DupliHood. Thus, I think the macro lens approach is a very valid one.

For colour negatives, two hints that helped me a lot with digitizing via camera:

1. Make a custom white balance by shooting through an unexposed piece of the negative. That gets you almost all the way to eliminating the orange hue. Even if shooting Raw (highly recommended) this gives you the best starting point.

2. If your Raw converter supports it, use a linear camera profile. The immediate output will look darker and flatter than you want, but it gives you the best control over whatever gamma/curves/contrast adjustments you want to make. Once you have found adjustments you like, if you have a large batch of images that are pretty much the same exposure you can use batch processing options to apply the same adjustments to each one and reduce the tedium.

Best wishes,
Sterling
--
Lens Grit

SterlingBjorndahl wrote:

For colour negatives, two hints that helped me a lot with digitizing via camera:

That is fortunately not something I often shot. It was mostly B&W negatives and color slides, with Cibachrome printing for the slides. For scanning, I have to admit that my CoolScan 9000 did better with color negatives because the scanning software has really good and easily-tweakable profiles for each of the common color films.

1. Make a custom white balance by shooting through an unexposed piece of the negative. That gets you almost all the way to eliminating the orange hue. Even if shooting Raw (highly recommended) this gives you the best starting point.

A good idea, although you'll still be losing DR, especially on the red channel and less so on the green (orange = red + a little green). Perhaps a better option would be to use a not-orange filter to increase captured DR? Of the most common color correction filters, I'd guess an 80B might work best, but an actual filter pack that precisely nulls the orange tint would be even better. I have a dichroic color head on my 23CII enlarger, so I could dial in the inverse color filtering using that as a copy light source. Even with a filter approximately nulling the orange cast, custom white balance is definitely better than the can-vary-with-each-slide auto setting for color negatives.

Note that I haven't personally tested color-correcting the light source for color negatives the above way, so who knows how well the theory really works?

2. If your Raw converter supports it, use a linear camera profile. The immediate output will look darker and flatter than you want, but it gives you the best control over whatever gamma/curves/contrast adjustments you want to make. Once you have found adjustments you like, if you have a large batch of images that are pretty much the same exposure you can use batch processing options to apply the same adjustments to each one and reduce the tedium.

Let's talk a bit about dynamic range (DR). Film response to light isn't linear to begin with, and it does some very funky things near min and max densities. Basically, color film might capture 11-12 stops of scene DR, but it turns into a negative or slide where only about 7-9 stops have approximately the right tones. It might take more than 15 stops of copy DR to digitally capture those "hidden" tones so that HDR tone mapping can make the details visible and approximately faithful to the original scene appearance. BTW, this seems to be largely a dye property, so metallic silver B&W images tend not to have as much hidden DR.

The happy result is that the in-camera multi-shot HDR on my Sonys does a surprisingly good job of recovering this hidden DR, commonly automatically recovering a couple of stops of DR that I couldn't even make out by direct viewing of the slide (humans only see 11-13 stops of DR). Combining a raw HDR sequence in postprocessing can improve image quality a tad more, but at least the tone mapping Sony uses for in-camera HDR JPEGs is pretty good without any fuss. Postprocessing raw HDR sequences would allow more tweaking of the color for color negative images, although I suspect you wouldn't lose too much inverting a tone-mapped JPEG taken through a filter like an 80B.

Alternatively to HDR sequences, single-shot DRO can help too, but usually isn't enough to get everything on my Sonys.

Particularly for HDR sequence captures, but in general when shooting with a copying attachment, don't forget to turn-off OIS/IBIS! You don't want the camera correcting for motions that the camera and film you're photographing make together.

Best wishes,
Sterling
--
Lens Grit

Sorry for my use of the American spelling of colour.

Thanks to everyone who replied. I’ll post a reply once I’ve made a decision. This before I buy anything I’ll give the tripod method a go. I’ve got all the gear I need and it would give me complete  flexibility.

Thanks for those insights about DR. It makes sense and actually may explain some anomalies I've seen regarding not being able to capture some shadow detail (the almost-clear spots) on negatives. I have a light source that has some temperature variability so I think I'll play around with that a bit. I think I have some gels around somewhere too. No 80B on hand but that's not hard to rectify.

I've tried HDR on some slides but was never happy with how the shadows looked. I could never get a result that matched the slide, despite the relatively modest DR of the source. Could well just be lack of experience.

Thanks,
Sterling
--
Lens Grit

ProfHankD wrote:

AceVentura0212 wrote:

Hello, all, hope this is the right place to post my question. I've been shooting digital for 20+ years but have recently acquired a few film cameras and now need to digitize my negatives. I've decided on using my Canon 5DsR (50mpx) camera and my Canon 100mm f/2.8 macro (the non-L) version. My question concerns the method of digitizing the negatives.

I have found on Amazon a Lens Tube mechanism that attaches to the lens and should work well with a light panel that I have.

This is most similar to what I use, which is the 3D-printed DupliHood typically on my Macro Rokkor 50mm f/3.5. Consider using your camera's HDR mode (if it has one). I'm not sure how the unit you link to would work with a 100mm macro -- I don't see much adjustment of the adapter's length.

I've gotten pretty close with a 100mm macro + FF. There are multiple 52mm extension tubes, and the slide/negative holder itself has some adjustability. I got it fairly recently so haven't started doing some serious scanning runs.

Alternatively, there is also the Copy Stand method which also seems to work well with a light panel. Click the two links to see the Amazon entries.

Can work too. Should be more versatile, but much harder to get everything aligned.

Which is why I eventually gave up on this approach, although I was using a DIY copy stand made from Home Depot parts and a Camvate superclamp.

So, as between the two is there a consensus of the better product?

In either case, don't have the light panel too close behind the negative, because that will help evenness of the illumination. Also be careful about shutter speed, because LED panels use PWM that can cause banding at some speeds.

This is some of the benefit of the JJC unit - it has a built-in diffuser, so if you separate it from your panel, you get even illumination even if your panel's diffuser is garbage.

There are a couple of articles which state that, theoretically, scanning with individual R, G, and B LED illumination in separate exposures that are then combined is best for film - https://medium.com/@alexi.maschas/color-negative-film-color-spaces-786e1d9903a4 for example.  The fact that Blackmagic's Cintel scanners operate in this manner strongly indicates there is merit to the approach.  I'm working on a setup that automatically does the R, G, and B exposures with a Bluetooth-controlled Neewer LED panel, then merges the R, G, and B CFA elements into a single DNG using some Python.

Finally, I have a Manfrotto tripod w a head that will tilt in such a way that I could conceivably use that instead.

The really slow to align option...

Been there done that, only do this if you enjoy pain and suffering.

Any thoughts and insights will be greatly appreciated.

I actually own a Nikon CoolScan 9000, which is one of the best dedicated scanners ever made, but easily get better quality using a Sony A7 with the DupliHood. Thus, I think the macro lens approach is a very valid one.

Both RT and darktable have built-in negative-inversion systems that are designed for raw inputs - https://discuss.pixls.us/t/any-interest-in-a-film-negative-feature-in-rt/12569 for the development history of RT's tool

Entropy512 wrote:

ProfHankD wrote:

In either case, don't have the light panel too close behind the negative, because that will help evenness of the illumination. Also be careful about shutter speed, because LED panels use PWM that can cause banding at some speeds.

This is some of the benefit of the JJC unit - it has a built-in diffuser, so if you separate it from your panel, you get even illumination even if your panel's diffuser is garbage.

Every diffuser has some texture, even if it's just dust, and copying slides/negatives tends to be stopped down enough to extend DoF. I found it better to REMOVE the built-in diffuser on my old Spiratone DupliScope (which is what my DupliHood is a hack on). YMMV.

There are a couple of articles which state that, theoretically, scanning with individual R, G, and B LED illumination in separate exposures that are then combined is best for film - https://medium.com/@alexi.maschas/color-negative-film-color-spaces-786e1d9903a4 for example. The fact that Blackmagic's Cintel scanners operate in this manner strongly indicates there is merit to the approach. I'm working on a setup that automatically does the R, G, and B exposures with a Bluetooth-controlled Neewer LED panel, then merges the R, G, and B CFA elements into a single DNG using some Python.

I wouldn't expect that to help... unless using a monochrome sensor rather than one with a CFA.

ProfHankD wrote:

I wouldn't expect that to help... unless using a monochrome sensor rather than one with a CFA.

Why not?  (Other than the resolution loss)

Use only the R sites when illuminated by R, the G sites when illuminated by G, and B sites when illuminated by B.  (e.g. throw away the G/B site data from the R-illuminated image, etc.)

Entropy512 wrote:

ProfHankD wrote:

I wouldn't expect that to help... unless using a monochrome sensor rather than one with a CFA.

Why not? (Other than the resolution loss)

Use only the R sites when illuminated by R, the G sites when illuminated by G, and B sites when illuminated by B. (e.g. throw away the G/B site data from the R-illuminated image, etc.)

All that you'd be doing is substituting one color filter for another, and I don't think quality of the CFA filters is the issue.

ProfHankD wrote:

Entropy512 wrote:

ProfHankD wrote:

I wouldn't expect that to help... unless using a monochrome sensor rather than one with a CFA.

Why not? (Other than the resolution loss)

Use only the R sites when illuminated by R, the G sites when illuminated by G, and B sites when illuminated by B. (e.g. throw away the G/B site data from the R-illuminated image, etc.)

All that you'd be doing is substituting one color filter for another, and I don't think quality of the CFA filters is the issue.

Huh? No I would not be. What color filter - LEDs are fundamentally monochromatic (unless you add a phosphor to eliminate this fundamental behavior), their color does not originate from a filter.  I'd be effectively eliminating the color filter behaviors completely.  Red photosites only lit by 625nm (640-650 is ideal, but 625 is much more common), green by 540nm, blue by 450.  Crosstalk between the filters is completely eliminated by throwing away the results of the G or B photosites when illuminated by 625nm and so on.

The reasoning for doing this is mentioned in the link I provided. Also see Nate Weatherly's discussion of why you want to avoid the orange mask wavelengths at https://discuss.pixls.us/t/digitizing-film-using-dslr-and-rgb-led-lights/18825/6 - this is done by using monochromatic light sources.

Yes, doing what Cintel does and using a monochrome sensor to get full R, G, and B data at each photosite would be ideal, but demosaicing a composited image generated from three separately illuminated shots should be close enough.

Entropy512 wrote:

ProfHankD wrote:

Entropy512 wrote:

ProfHankD wrote:

I wouldn't expect that to help... unless using a monochrome sensor rather than one with a CFA.

Why not? (Other than the resolution loss)

Use only the R sites when illuminated by R, the G sites when illuminated by G, and B sites when illuminated by B. (e.g. throw away the G/B site data from the R-illuminated image, etc.)

All that you'd be doing is substituting one color filter for another, and I don't think quality of the CFA filters is the issue.

Huh? No I would not be. What color filter - LEDs are fundamentally monochromatic (unless you add a phosphor to eliminate this fundamental behavior), their color does not originate from a filter. I'd be effectively eliminating the color filter behaviors completely. Red photosites only lit by 625nm (640-650 is ideal, but 625 is much more common), green by 540nm, blue by 450. Crosstalk between the filters is completely eliminated by throwing away the results of the G or B photosites when illuminated by 625nm and so on.

The reasoning for doing this is mentioned in the link I provided. Also see Nate Weatherly's discussion of why you want to avoid the orange mask wavelengths at https://discuss.pixls.us/t/digitizing-film-using-dslr-and-rgb-led-lights/18825/6 - this is done by using monochromatic light sources.

Yes, doing what Cintel does and using a monochrome sensor to get full R, G, and B data at each photosite would be ideal, but demosaicing a composited image generated from three separately illuminated shots should be close enough.

Yes, LEDs are relatively pure wavelength emitters (about 10nm bands). However, for three narrow bands of lighting, the crosstalk is a directly computable ratio, so with the right raw processing, a single exposure can still suffice. Again, all you're doing is effectively sharpening the CFA filtering. In fact, wisely picking the 3 LED wavelengths, you probably wouldn't even need to correct for crosstalk -- after all, the goal is really just to get rid of the orange mask.

ProfHankD wrote:

Entropy512 wrote:

ProfHankD wrote:

Entropy512 wrote:

ProfHankD wrote:

I wouldn't expect that to help... unless using a monochrome sensor rather than one with a CFA.

Why not? (Other than the resolution loss)

Use only the R sites when illuminated by R, the G sites when illuminated by G, and B sites when illuminated by B. (e.g. throw away the G/B site data from the R-illuminated image, etc.)

All that you'd be doing is substituting one color filter for another, and I don't think quality of the CFA filters is the issue.

Huh? No I would not be. What color filter - LEDs are fundamentally monochromatic (unless you add a phosphor to eliminate this fundamental behavior), their color does not originate from a filter. I'd be effectively eliminating the color filter behaviors completely. Red photosites only lit by 625nm (640-650 is ideal, but 625 is much more common), green by 540nm, blue by 450. Crosstalk between the filters is completely eliminated by throwing away the results of the G or B photosites when illuminated by 625nm and so on.

The reasoning for doing this is mentioned in the link I provided. Also see Nate Weatherly's discussion of why you want to avoid the orange mask wavelengths at https://discuss.pixls.us/t/digitizing-film-using-dslr-and-rgb-led-lights/18825/6 - this is done by using monochromatic light sources.

Yes, doing what Cintel does and using a monochrome sensor to get full R, G, and B data at each photosite would be ideal, but demosaicing a composited image generated from three separately illuminated shots should be close enough.

Yes, LEDs are relatively pure wavelength emitters (about 10nm bands). However, for three narrow bands of lighting, the crosstalk is a directly computable ratio, so with the right raw processing, a single exposure can still suffice. Again, all you're doing is effectively sharpening the CFA filtering. In fact, wisely picking the 3 LED wavelengths, you probably wouldn't even need to correct for crosstalk -- after all, the goal is really just to get rid of the orange mask.

Unless you have a camera with its IR blocking filter removed. Then you can add a 4th set of LEDs, IR, to the sequence. High end film scanners used to do this because film dyes are mostly transparent to infrared, so you can get a "dust map" from an IR image.

Run the four images through VueScan, which does support this sort of thing, and you don’t have to do any coding.

Joseph S Wisniewski wrote:

ProfHankD wrote:

Yes, LEDs are relatively pure wavelength emitters (about 10nm bands). However, for three narrow bands of lighting, the crosstalk is a directly computable ratio, so with the right raw processing, a single exposure can still suffice. Again, all you're doing is effectively sharpening the CFA filtering. In fact, wisely picking the 3 LED wavelengths, you probably wouldn't even need to correct for crosstalk -- after all, the goal is really just to get rid of the orange mask.

Unless you have a camera with its IR blocking filter removed.

Actually, even with it present you still get enough sensitivity if you only have NIR lighting. Typical NIR-blocking filters only cut sensitivity by 7-9 EV, but cameras now often have >13 EV DR, so you even see traces of NIR leakage in normal use (e.g., that's most of how you get PF).

Then you can add a 4th set of LEDs, IR, to the sequence. High end film scanners used to do this because film dyes are mostly transparent to infrared, so you can get a "dust map" from an IR image.

In fact, a few cameras use four-color CFAs, in which case you can solve for R, G, B, NIR . I did that with ordinary lighting, but it will work even better with narrow-band emitters like LEDs. BTW, it's more of a defect map -- a lot of what you see in NIR is the scratches, not just dust. Scratches are slightly harder to repair well because they affect numbers of "spatially correlated" pixels, whereas dust tends to be just diffuse spots.

Run the four images through VueScan, which does support this sort of thing, and you don’t have to do any coding.

Yup. Hamrick's VueScan is excellent in many ways. That said, I'm not sure he repairs scratches the right way -- almost nobody does. The right way is to interpolate from the neighbors AND THEN ADD BACK NOISE to match the statistics for similar regions of the image. For example, I did that in KARWY-SR for credibly repairing the Sony sensor PDAF striping problem...