Why doesn't anyone make this?

[MattKing]

Earlier in this thread there was a question about building a template or jig for camera scanning and stitching. Here's an example I wrote about in the context of 4x5 negatives: https://www.largeformatphotography....era-scanning-on-the-cheap-an-example-approach This approach can easily be adapted to any other film size where multiple frames are needed.

This approach is easy to use and vastly faster than a flatbed scan on an Epson V750. However, I'm going to make a new one to reduce the number of frames, and the resulting resolution. My current setup gives me 2,666 ppi, which is overkill for my needs. The files are huge, and things slowdown too much in Lightroom. The next version I'm going to build will have a target resolution of 2,000 ppi, which I can achieve with 6 frames instead of the 12 needed for 2,666 ppi. For the rare cases where 6 frames do not allow enough overlap, I'll be designing-in the ability to shoot 9 to maximize overlap. 2,000 ppi is still more resolution than I need for my typical largest size print (16" on the short edge), but the resulting files are quite manageable, so it's not a problem to have "too much" resolution.

Thanks for this.
The only frustrating thing is that you have to be a registered member of the Large Format forum in order to see anything more than thumbnail versions of the illustrations.
I probably should go through the process of registering (and waiting for acceptance) there, even though I don't shoot large format.

 

[rdeloe]

Matt, the membership on LFP is free, and it's worth it even if you're not shooting large format. There are some good threads on digitizing film using scanners and cameras. Some of the forum members have made a major investment in figuring things out.

Thanks for this.
The only frustrating thing is that you have to be a registered member of the Large Format forum in order to see anything more than thumbnail versions of the illustrations.
I probably should go through the process of registering (and waiting for acceptance) there, even though I don't shoot large format.

 

[PhilBurton]

In my opinion:

1. PS is time consuming.
2. Standard lenses don't have enough "flatness" of the image to attain good sharpness and the detail when photographing the negatives and macro lenses are, well, pricey.

I tried to "scan" a negative via digital camera and encountered a few downsides: first of all, I had to grab a macro lens. Then, I had to stop it down. A lot. Next step was achieving correct white balance in PS. Also, medium format shots still laked sharpness, no matter what I tried. If I scanned it piece by piece, it took even more time and that's where I dropped the whole thing and went straight to flatbed scanner. Silverfast made my life easier and now I have enough time to read a book while my negatives are batch-scanned.

What features in Silverfast allow you to effectively batch scan negatives?

Out of curiosity, did you compare Vuescan to see if it had similar features?

 

[blacksquare]

Can someone post a example of high res DSLR "scan" of 120 negative? At best stitched method?
When I tried it with Fuji X-T1 and Canon FD 50, I got a lot of extra data but at 1:1 only very little extra detail.

 

[rdeloe]

In my testing, I saw no real benefits to a resolution of 3,000 ppi compared to 2,666 ppi. As you say, more data but not more detail... I performed this test using two different lenses on a Fuji X-T2: the Olympus OM 90/2 and the Olympus OM 80/4 Macro. I'm sure there are better macro lenses out there, but these are both exceptionally good.

How much resolution do you need? Let's say you have a 6x7 negative. You can cover that with two overlapping frames at 2.26" x 1.57", but you should use a third one to guarantee enough overlap. This will give you a resolution of 2,074 ppi using your Fuji X-T1 -- which works out to 5,716 x 4,895 pixels or 28 MP on a 6x7 negative.

By increasing the magnification you can increase resolution. However, you should keep in mind that the smaller the image area of each section you have to stitch together, the greater the likelihood that you will run into pictures that won't stitch unless you use a highly accurate jig (or some kind of X-Y stage), and then stitch the images together using templates in PT GUI (or a similar tool). I've run into negatives that have large areas of even tone that wouldn't stitch using Lightroom, Photoshop, Microsoft ICE, or PT GUI. Unfortunately, using my current setup, I am unable to achieve the necessary precision to use templates in PT GUI (see above).

Can someone post a example of high res DSLR "scan" of 120 negative? At best stitched method?
When I tried it with Fuji X-T1 and Canon FD 50, I got a lot of extra data but at 1:1 only very little extra detail.

 

[alanrockwood]

Is anyone here using Hugin for the stitching in this application?

 

[M-88]

What features in Silverfast allow you to effectively batch scan negatives?

The "batch scan" feature itself?

Out of curiosity, did you compare Vuescan to see if it had similar features?

Yes, I compared and it's trickier with Vuescan. Setting/adjisting frames takes more time, simple zooming in and also out requires re-run of pre-scan (not sure if this can be avoided somehow). One thing Vuescan got right, that Silverfast did not, personally for me is dust/scratch removal.

 

[PhilBurton]

One thing Vuescan got right, that Silverfast did not, personally for me is dust/scratch removal.

Good to know, because Silverfast makes a big deal on their website about dust/scratch removal. B&W or color negs?

Phil

 

[M-88]

Good to know, because Silverfast makes a big deal on their website about dust/scratch removal. B&W or color negs?

Phil

BW and color as well.

I must also mention that I don't have the latest version of SF, only version 6.something. And perhaps things are improved in version 8. As for Vuescan, both that I've used - version 8 and 9 had dust removal via second pass of scanner, must have been infrared or something, because it does a wonderful job. If Vuescan had proper batch scan, I wouldn't be using SF at all.

 

[Helge]

If you really want to make 4000 dpi the minimum scan resolution you’re willing to accept, I’m not going to disagree as only can decide what is acceptable for you, however, I own and operate a film lab and process and scan *a lot* of film. 24MP shows film grain in most emulsions in 120 format. It’s a simple fact. Once you’re seeing film grain, more resolution is diminishing returns.

Is it grain, or grain aliasing you are seeing? Also resolution doesn't come to a dead stop at the grain level.
I have scanned Ektar and Portra up to 80 MP resolutions and seen visible improvements. Diminishing returns sure, but you go from a face in the distance that is hardly recognisable, to quite clear, or text and other details that goes from a blurry mess, to just readable.
There is grain/dye clusters "on top" but some parts of the film has picked up the detail.

Colour resolution, that is one of films many strengths, is also markedly improved with higher resolutions. The effect of the bayer demosaicing is diminished to a great degree.

I've also experimented with attenuating the exposure and light over the same part, to get all the dynamic range of normal or thick negatives, and the results are promising.

Is it worth it for Instagram or Flickr? Probably not.
But the minute you want to go exploring around on the image, not even pixel peeping, or crop or print in sizes approaching poster format, it does pay off.

 

[Adrian Bacon]

Is it grain, or grain aliasing you are seeing? Also resolution doesn't come to a dead stop at the grain level.
I have scanned Ektar and Portra up to 80 MP resolutions and seen visible improvements. Diminishing returns sure, but you go from a face in the distance that is hardly recognisable, to quite clear, or text and other details that goes from a blurry mess, to just readable.
There is grain/dye clusters "on top" but some parts of the film has picked up the detail.

Colour resolution, that is one of films many strengths, is also markedly improved with higher resolutions. The effect of the bayer demosaicing is diminished to a great degree.

I've also experimented with attenuating the exposure and light over the same part, to get all the dynamic range of normal or thick negatives, and the results are promising.

Is it worth it for Instagram or Flickr? Probably not.
But the minute you want to go exploring around on the image, not even pixel peeping, or crop or print in sizes approaching poster format, it does pay off.

For color or black and white?

EDIT: for color, the debayering process smears quite a bit of fine detail, so there really isn't such a thing as grain aliasing with a CFA sensor, so it’s essentially the same as taking the original picture with a DSLR, you essentially have a CFA image that you have to do some post processing on to get an RGB image. The difference is you effectively have the worlds most awesome film simulation because you're actually shooting it on film. In those instances, more resolution absolutely is better, though I’d rather have a higher resolution CFA sensor and get more res in a single capture than do the stitch route. For my scan process, all my code does the inversion, linearization, and color space conform on the raw bayer samples, and what gets fed in LR is a linear floating point bayer array. All of the spatial image detail that made it onto the sensor from the film is still there and intact. You’d be amazed at how much spatial detail gets eaten up if you debayer a scan, then do all your post processing to get a color positive as opposed to doing all your post to get a color positive on the raw bayer array then feeding the color positive bayer array to LR to debayer. You also get a lot less color pollution between the channels because your negative’s red and green and blue values aren’t being interpolated into adjacent pixels to make an RGB image before you try to invert and remove the color mask and adjust the contrast of each channel. It’s way better to do all that before you debayer.

For black and white, since I’m taking an image of something that has no color information, the only color information that is hitting the bayer array is from my light source. I white balance that out on the raw bayer samples and proceed to treat the bayer array directly as a monochrome image. In that instance, I really am sampling 6000+ samples over the width of a 36mm frame and 4000+ samples over the 24mm height for 135 film because there is absolutely no debayering going on. What gets fed into LR is a linear floating point monochrome image with all of the spatial information that hit the bayer sensor intact.

When people scan their black and white film with a DSLR and marvel about the huge image quality boost over a flatbed, they’re seeing an image that has been white balanced and debayered. 50% of a bayer arrays sensels are dedicated to red and blue, so when you white balance and then debayer, you’re polluting 50% of your captured spatial information through the debayering process to arrive at rgb values of an image that doesn’t actually have RGB information in it. Stop and noodle on that for a minute.... And people are declaring the minimum acceptable resolution they’re willing to accept based on that and come up with ever more complex ways to do the initial physical scan to try to get more resolution?

Umm... OK... if that’s really the route you want to go, then by all means, don’t let me stop you. How you handle the raw sample data makes a really big difference in the output. I’d rather have a reasonably high resolution source and optimize how I handle that before making any judgment calls about whether to either go get more raw sensor resolution, or do other things like scan and stitch.

The same goes for dynamic range of the film. The film is effectively gamma encoded. The actual density range fits well inside what a modern DSLR can capture in a single capture. How you reverse that gamma to get linear makes a big difference.

 

[PhilBurton]

For color or black and white?

EDIT: for color, the debayering process smears quite a bit of fine detail, so there really isn't such a thing as grain aliasing with a CFA sensor, so it’s essentially the same as taking the original picture with a DSLR, you essentially have a CFA image that you have to do some post processing on to get an RGB image. The difference is you effectively have the worlds most awesome film simulation because you're actually shooting it on film. In those instances, more resolution absolutely is better, though I’d rather have a higher resolution CFA sensor and get more res in a single capture than do the stitch route. For my scan process, all my code does the inversion, linearization, and color space conform on the raw bayer samples, and what gets fed in LR is a linear floating point bayer array. All of the spatial image detail that made it onto the sensor from the film is still there and intact. You’d be amazed at how much spatial detail gets eaten up if you debayer a scan, then do all your post processing to get a color positive as opposed to doing all your post to get a color positive on the raw bayer array then feeding the color positive bayer array to LR to debayer. You also get a lot less color pollution between the channels because your negative’s red and green and blue values aren’t being interpolated into adjacent pixels to make an RGB image before you try to invert and remove the color mask and adjust the contrast of each channel. It’s way better to do all that before you debayer.

For black and white, since I’m taking an image of something that has no color information, the only color information that is hitting the bayer array is from my light source. I white balance that out on the raw bayer samples and proceed to treat the bayer array directly as a monochrome image. In that instance, I really am sampling 6000+ samples over the width of a 36mm frame and 4000+ samples over the 24mm height for 135 film because there is absolutely no debayering going on. What gets fed into LR is a linear floating point monochrome image with all of the spatial information that hit the bayer sensor intact.

When people scan their black and white film with a DSLR and marvel about the huge image quality boost over a flatbed, they’re seeing an image that has been white balanced and debayered. 50% of a bayer arrays sensels are dedicated to red and blue, so when you white balance and then debayer, you’re polluting 50% of your captured spatial information through the debayering process to arrive at rgb values of an image that doesn’t actually have RGB information in it. Stop and noodle on that for a minute.... And people are declaring the minimum acceptable resolution they’re willing to accept based on that and come up with ever more complex ways to do the initial physical scan to try to get more resolution?

Umm... OK... if that’s really the route you want to go, then by all means, don’t let me stop you. How you handle the raw sample data makes a really big difference in the output. I’d rather have a reasonably high resolution source and optimize how I handle that before making any judgment calls about whether to either go get more raw sensor resolution, or do other things like scan and stitch.

The same goes for dynamic range of the film. The film is effectively gamma encoded. The actual density range fits well inside what a modern DSLR can capture in a single capture. How you reverse that gamma to get linear makes a big difference.

This is very convincing, that the scan output should be a RAW file instead of a JPG or TIFF. Now, how can I do that with my Nikon 5000?

 

[Adrian Bacon]

This is very convincing, that the scan output should be a RAW file instead of a JPG or TIFF. Now, how can I do that with my Nikon 5000?

In my case, I wrote a bunch code to do it. Black and white is pretty straightforward, color is a whole ‘nother animal.

I get requests to make my code available all the time (usually for free), but given that I have a fair amount time and energy invested in it and it’s geared towards volume scanning in a lab environment, I’ve chosen to keep it in house for the time being.

 

[alanrockwood]

Adrian,

How much blurring occurs via the anti-aliasing filter? I realize that this must depend on the make/model of camera, but can some general comments apply?

As you probably know, but for the benefit of those who don't, if the anti-aliasing filter was a true "brick wall" filter, in the sense that the signal processing people use the term, there should be no moire effect in images, but most digital cameras are subject to moire effects under certain conditions, so I assume that the ant-aliasing filters are very imperfect.

Also in post #61 you strongly hinted at an important point that is often not fully understood by the camera-using public. Let me amplify the point using an example. Because the image sensor in most digital cameras is a Bayer sensor, a 12 megapixel camera is actually not a 12 megapixel camera but instead is a 6 megapixel green camera, a 3 megapixel blue camera, and a 3 megapixel red camera, with the color channels interleaved. This fact sets upper limits on the achievable resolution, and that limit is less than what one would infer if the camera were a true 12 megapixel camera. The de-Bayering process has to fill in the missing information. For example, it must guess at how much red to assign to a pixel position that only senses a green. The mathematical guesswork that goes into this interpolation process can never truly supply the missing information without the possibility of error, though the guesses might be pretty good for most images most of the time. Anyway, as you suggested, one of the effects is that there is a loss of resolution, and this lost resolution cannot be restored without the possibility of error. You also discuss some interesting issues of color pollution between channels resulting from the fact that the sensors are Bayer sensors.

By the way, Foveon-type sensors partially overcome some of these problems, but introduce other problems of their own.

 

[Adrian Bacon]

How much blurring occurs via the anti-aliasing filter? I realize that this must depend on the make/model of camera, but can some general comments apply?

That’s hard to say... they don’t have to be perfect, just enough to start rolling the detail off a little before the Nyquist. You don’t have to completely get rid of it, just attenuate enough that it’s not really going to cause issues.

I’ve shot Canon for most of my adult life and Canon lately has been bagged on quite a bit by reviewers for including an optical low pass filter in the optical path where most everybody else has been forgoing one. They are not dummies and they include one for a reason. I almost never see moire with my canon gear, and I would be very disappointed to see a shoot hosed up because moire popped up, especially if it was a high dollar shoot. Even a modest entry level DSLR has enough raw resolution for most commercial print purposes, so the whole “extra detail because there’s no low pass filter” is really for the pixel peepers that think that’s going to make a difference in a reasonable print or magazine spread. By the time the image gets resized down to print resolution and sharpened for that output, that extra detail is moot, but... you will see the moire...

 

[Grim Tuesday]

I think one thing that everyone in this thread is missing is that a jig or outfit for doing this is totally not necessary. Just push the neg carrier a little bit, take a picture, push the neg carrier a little bit, take a picture and so on and so forth. Takes about 10 minutes to scan a roll of 120. I have a script to do photomerges in Photoshop, which take a while but do not require user input. Works great with 6 pictures for 6x6 film. 9 out of 10 times, Photoshop has no trouble stitching everything up, even if they're very uneven. The bigger issue for setups with stitched negatives is corner sharpness of your macro lens, which often degrades as magnification is increased due to field flatness issues. I used a Nikkor 55/3.5 for a while and there were obvious zones of sharpness in my resulting stitched image. I upgraded to a 55/2.8 and it is much better, though still not perfect. Either one is massively better than my Epson V700. I consult Coinimaging for my macro lens testing needs: http://coinimaging.com/nikon_55microais.html

That said, I like the idea of the 3D printer based motorized jig for this. If I could put 3 negatives in a carrier and have it scanned by pushing a button, I would throw away my V700.