grayscale or rgb for b+w negs

[alanrockwood]

Alan It would be helpful if you could list which films require 16 bit? For example, I use Tmax 100, Tmax 400, Velvia 50 and Provia 100. These are very fine relatively grain-free. I normally scan 16 bit. I use Lightroom in post. What are your recommendations?

I can't give a specific recommendation. I did make some theoretical calculations, based on reported granularity of TMAX-100, and within the reasonable density range it looks like 8 bit scanning will probably work, but it might be a bit "iffy". The best thing to do would be actual testing, which I have not done.

There are several approaches one could use for testing. One would be to create a target with a smooth gradient. Scan it with 8 bit and 16 bit scans. Then see if you can see banding in the result. You could then do some extreme photoshop manipulations on those images to see if banding or other artifacts show up on the 8 bit scans that are not present in the 16 bit scans. Be sure to convert any image to 16 bit before doing any manipulations.

Another would be to shoot a very smooth target that has no gradient. Shoot it at several exposure levels. Be sure to defocus the lens to make sure there is no detail showing up in the smooth target image. Scan the image using 8 bit scanning and look at the histogram of distribution of intensity levels in very small segments near the center of the image. If there is more than one level in the histograms of those small segments it means that there is some noise in the image (from grain and from other sources), and that the noise is comparable to or greater than the ADC step size. In that case 8 bit scanning is probably sufficient. It is important to look at small segments, and not large segments or the whole image. I can explain why if you want, but I won't do the explanation in this post.

 

[Alan Edward Klein]

I can't give a specific recommendation. I did make some theoretical calculations, based on reported granularity of TMAX-100, and within the reasonable density range it looks like 8 bit scanning will probably work, but it might be a bit "iffy". The best thing to do would be actual testing, which I have not done.

There are several approaches one could use for testing. One would be to create a target with a smooth gradient. Scan it with 8 bit and 16 bit scans. Then see if you can see banding in the result. You could then do some extreme photoshop manipulations on those images to see if banding or other artifacts show up on the 8 bit scans that are not present in the 16 bit scans. Be sure to convert any image to 16 bit before doing any manipulations.

Another would be to shoot a very smooth target that has no gradient. Shoot it at several exposure levels. Be sure to defocus the lens to make sure there is no detail showing up in the smooth target image. Scan the image using 8 bit scanning and look at the histogram of distribution of intensity levels in very small segments near the center of the image. If there is more than one level in the histograms of those small segments it means that there is some noise in the image (from grain and from other sources), and that the noise is comparable to or greater than the ADC step size. In that case 8 bit scanning is probably sufficient. It is important to look at small segments, and not large segments or the whole image. I can explain why if you want, but I won't do the explanation in this post.

No that's OK. I'll just keep scanning at 16 bit and saving as a tiff since who knows what film and when banding might show up. Wouldn't the size of the print also play an effect? Another issue is the type of post processing program. While Photoshop keeps 16 bit, Elements does not. Also, once you store in jpeg, you reduce to 8 bit (I think). I suppose, as long as memory isn't an issue or processing speed, it might be better to scan at 16 bit and save in tiff.

 

[lenshood]

Generally there be one of the channels that gives the sharpest image. The preferred channel might depend on the brand of scanner. As I recall, for my canon FS4000us scanner it is the green channel.

forgive, please, the first of what are bound to be many elementary questions: how does one determine/discover which channel gives the sharpest image for a particular scanner?

 

[lenshood]

Let me refine this concept just a bit more. If you scan in 8 bit then whenever you start doing digital image processing you should always convert the image to 16 bit and keep it at 16 bit all the way through the process, including any intermediate saves and the final save. This is to eliminate issues due to roundoff error.

more elementary questions:

first, I want to make sure I understand this suggestion: would this mean scanning in 24 bit rgb, saving from the scanning software as a tiff, and then converting to 48 bit in LR or PS or what have you, and then removing the two unwanted color channels? I’m pretty certain I’m missing something basic here, so apologies.

 

[Ko.Fe.]

Sorry. I got confused with statements others have made to scan in color and then use the green portion for better sharpness. I was trying to understand how to do that with Lightroom.

Lightroom doesn’t scan. You could get more details only during scanning.
I have tried green only channel for scanning and didn’t find it to be any better.

 

[alanrockwood]

No that's OK. I'll just keep scanning at 16 bit and saving as a tiff since who knows what film and when banding might show up. Wouldn't the size of the print also play an effect? Another issue is the type of post processing program. While Photoshop keeps 16 bit, Elements does not. Also, once you store in jpeg, you reduce to 8 bit (I think). I suppose, as long as memory isn't an issue or processing speed, it might be better to scan at 16 bit and save in tiff.

Yes, there is nothing wrong with your strategy.

 

[alanrockwood]

more elementary questions:

first, I want to make sure I understand this suggestion: would this mean scanning in 24 bit rgb, saving from the scanning software as a tiff, and then converting to 48 bit in LR or PS or what have you, and then removing the two unwanted color channels? I’m pretty certain I’m missing something basic here, so apologies.

Yes, that is one way to do it.

I think some scanning systems (hardware plus software ) will also let you choose which channel to save for B&W, so that is another route to the same end, i.e. scan in 8 bit, saving the channel of choice. Of course there is nothing wrong with scanning in 16 bit, other than some storage cost, which is practically free these days anyway. It's just that if certain criteria are met you really don't lose anything significant by scanning in 8 bit.

 

[grat]

There is an opensource tool called "ImageMagick" (Linux/OSX/iOS/Windows), which among other things, can separate images by color channel into multiple files.

 

[138S]

Alan It would be helpful if you could list which films require 16 bit?

Alan, we have to scan and edit 16bits/channel always, and saving with file a format that conserves the 16/bits, this is TIFF. If not you may suffer banding in the edition, and you may loss shadow detail.

The 16 bits/channel is more important when film records an ample dynamic range, not mattering what kind of film.

 

[hsandler]

Way way back I compared scanning a sharp b&w negative in 16 bit grayscale vs. 48 bit colour, both with Epsonscan at 4800 ppi on a V700 at optimal height, and then checking if the green or another colour channel was sharper. I didn’t see any difference or at least not enough to bother with the colour channel workflow. I understand the theory as to why one colour might be sharper—if the scanner lens focuses different wavelengths slightly differently there could be a difference.

 

[alanrockwood]

Alan, we have to scan and edit 16bits/channel always, and saving with file a format that conserves the 16/bits, this is TIFF. If not you may suffer banding in the edition, and you may loss shadow detail.

The 16 bits/channel is more important when film records an ample dynamic range, not mattering what kind of film.

There is nothing wrong with scanning in 16 bits, but if the grain or other sources of noise in the system are greater than or roughly equal to the step size of an 8 bit ADC then you won't get banding or loss of shadow detail if you use an 8 bit digitizer. It is basically equivalent to "dithering" a technique that is well known in the digital signal processing world. It is a method to eliminate artifacts due to ADC step size. Banding in an image is an example of an artifact arising from ADC step size.

Dithering often involves artificially adding noise to the signal. In those applications you make a trade-off. You accept a little bit of extra noise in exchange for eliminating ADC artifacts. However, if the signal is already noisy (e.g. grain in the negative, or electronic noise in the photodiodes in the scanner's sensors) then you don't even have to make the trade-off because you don't have to artificially add any noise.

I had a whole thread on this where I explained it in detail. I also posted some images obtained from another source that demonstrated the principle.

Let me emphasize that there is nothing mysterious or magic about this. It is entirely based on well established and well-known principles in the digital signal processing world.

Please also note that in my earlier post I also said that if you are doing image manipulation then you should always convert an 8 bit image to a 16 bit image before doing the manipulations.

 

[138S]

but if the grain or other sources of noise in the system are greater than or roughly equal to the step size of an 8 bit ADC then you won't get banding or loss of shadow detail if you use an 8 bit digitizer.

In general, 8bits is uncacceptable in a Pro edition.

Anyway it depends on the dynamic range in the film, the curve edition you require, and how careful you are in the edition. A moderate curve edition in 8 bits may not have visible harmful effects, but with 8 bits problems come easy.


> if you have to pull deep shadows then you may have problems, those deep shadows can have good detail, but at 8 bits you only have 256 levels and those shadows may be encoded in 8 gray levels (specially in a high dynamic range negative), and as you pull the shadows you loss the intermediate levels. Say you locally pull some shadows from 8 level to 64 level, after the expansion you will have only the levels 0, 8, 16, 24... with nothing in the middle...


> in lightroom, with 8 bit if you cascade several curve editions with expansions-compresions you easily get a degradation, if you use non destructive adjustment layers in Photoshop you don't have that problem, but you have to operate with layers.


> You may have to cascade several different sharpening operations of different radius, if working 16 bits you get an accurate job, with 8 bits those operations can be very destructive, in the deep shadows specially.


We may use 8 bit workflow for a moderate correction in a not much important image, but for a proficient edition we have to operate an oversampled image at 16 bits. Today's computers allow that, we have many GB in the RAM and M.2 disks are x30 faster than magnetic disks, also a 30,000 passmark cpu is $200 only. In the past it was very slow to operate with big images so many sacrifices had to be allowed and wisdom was required to get the optimal point, today this has changed.

 

[alanrockwood]

There is nothing wrong with scanning in 16 bits, but if the grain or other sources of noise in the system are greater than or roughly equal to the step size of an 8 bit ADC then you won't get banding or loss of shadow detail if you use an 8 bit digitizer. It is basically equivalent to "dithering" a technique that is well known in the digital signal processing world. It is a method to eliminate artifacts due to ADC step size. Banding in an image is an example of an artifact arising from ADC step size.

Dithering often involves artificially adding noise to the signal. In those applications you make a trade-off. You accept a little bit of extra noise in exchange for eliminating ADC artifacts. However, if the signal is already noisy (e.g. grain in the negative, or electronic noise in the photodiodes in the scanner's sensors) then you don't even have to make the trade-off because you don't have to artificially add any noise.

I had a whole thread on this where I explained it in detail. I also posted some images obtained from another source that demonstrated the principle.

Let me emphasize that there is nothing mysterious or magic about this. It is entirely based on well established and well-known principles in the digital signal processing world.

Please also note that in my earlier post I also said that if you are doing image manipulation then you should always convert an 8 bit image to a 16 bit image before doing the manipulations.

I know that some won't believe what I am saying. If anyone would like to prove me wrong and would like to do a little experiment, here's what you can do.

Find or create a negative of an object with a smooth gradient. Scan the negative twice with a high resolution film scanner, like a 4000 dpi Nikon or Canon film scanner. For one scan do it in 8 bit mode, saving as a tiff. For the other scan do it in 16 bit mode and save it as a tiff.

Read the two negatives into photoshop, or whatever your favorite image processing application is. Before you do anything else, convert the 8 bit image to 16 bits. From now on keep that image in 16 bit format, including any saves. From now on you will be treating the two images in exactly the same way.

Now do some image manipulation on the two images, the one that was converted to 16 bits and the one that was created in 16 bits. Do the manipulation using the histogram correction method. That way you can do the exact same manipulations to the two images.

Are you able to produce banding in one but not the other image? Can you even see a difference between the two manipulated images by visual inspection?

Try it with Tri-X film, then FP-4 plus, then Tmax 100.

My prediction is that you won't be able to produce banding with Tri-X or FP-4 plus, and probably not with Tmax 100.

Something like Velvia might give a different result, but Velvia isn't a negative film.

Now for a related thought. Have any of you ever scanned a negative (any negative) using a high dpi scanner in 8 bit mode, and then converted it to 16 bit mode, keeping it in 16 bit mode from then on, and then been able to produce banding by manipulating the image?

 

[alanrockwood]

In general, 8bits is uncacceptable in a Pro edition.

Anyway it depends on the dynamic range in the film, the curve edition you require, and how careful you are in the edition. A moderate curve edition in 8 bits may not have visible harmful effects, but with 8 bits problems come easy.


> if you have to pull deep shadows then you may have problems, those deep shadows can have good detail, but at 8 bits you only have 256 levels and those shadows may be encoded in 8 gray levels (specially in a high dynamic range negative), and as you pull the shadows you loss the intermediate levels. Say you locally pull some shadows from 8 level to 64 level, after the expansion you will have only the levels 0, 8, 16, 24... with nothing in the middle...


> in lightroom, with 8 bit if you cascade several curve editions with expansions-compresions you easily get a degradation, if you use non destructive adjustment layers in Photoshop you don't have that problem, but you have to operate with layers.


> You may have to cascade several different sharpening operations of different radius, if working 16 bits you get an accurate job, with 8 bits those operations can be very destructive, in the deep shadows specially.


We may use 8 bit workflow for a moderate correction in a not much important image, but for a proficient edition we have to operate an oversampled image at 16 bits. Today's computers allow that, we have many GB in the RAM and M.2 disks are x30 faster than magnetic disks, also a 30,000 passmark cpu is $200 only. In the past it was very slow to operate with big images so many sacrifices had to be allowed and wisdom was required to get the optimal point, today this has changed.

I said nothing about 8 bit work flow. I only dealt with 8 bit scanning. To meet the protocol I specified the image needs to be converted to 16 bit mode before anything else is done, and it must stay in 16 bit mode for all subsequent operations.

Why does it matter? Two reasons, one trivial and the other not. The trivial one is that it takes half the storage space to store an 8 bit scan compared to a 16 bit scan. With hard disk space so cheap these days that reason is probably trivial. The one that is not trivial is that certain hardware/software scanner systems only allow 8 bit scans. Leafscan operated under certain software systems is an example. This is a non-trivial example, though it is an uncommon example because Leaf scanners are not so common these days.

 

[138S]

I only dealt with 8 bit scanning.

If the scene is dull (a shot in the shadow) then 8 bits scanning will have little drawbaks, if you don't want to pull shadows much.


But if you have strong highlights and deep shadows in the same scene then you cannot take all with 8 bits, you have to compress highlights and shadows in the scanning if you don't want to clip, so you don't keep your choices open for the edition.

With a 16 bits scanning, in the scan, you have enough grays levels in the shadows and in the highlights to pull detailed textures there, with 8 bits you only can encode well the mids and you cannot encode well shadows/highlights, so in the edition you are tied to the compressions made in the scanning, so you don't keep your choices open for the edition.

The trivial one is that it takes half the storage space to store an 8 bit scan compared to a 16 bit scan. .... image needs to be converted to 16 bit mode before anything else is done

This is contradictory... You save space scanning 8 bits and later you need the space anyway to edit. You have to archive the edited 16 bits image anyway, for the case you want to make modifications in the future,

if the image is not important you may delete the 16 bits after edition, so the additional 16 bits image space usage is temporary, and if the image is important you will want to conserve the 16 bits version anyway.

certain hardware/software scanner systems only allow 8 bit scans.

The cheap Epson scanners allow 16 bits/channel, even the entry level V500. If a system does not allow 16 bits then simply avoid it: it does not allow a proficient edition for contrasty shots.

 

[Billy Axeman]

....
Now for a related thought. Have any of you ever scanned a negative (any negative) using a high dpi scanner in 8 bit mode, and then converted it to 16 bit mode, keeping it in 16 bit mode from then on, and then been able to produce banding by manipulating the image?

Thanks Alan for your insights.

I sometimes have a (B&W) photo with a blown out sky, which I replace with an artificial gradient (Graduated filter in Paint Shop Pro), which runs from black to white, with Opacity=10 (or about that).
When the photo was scanned to 8 bit JPEG, the gradient shows banding. When I convert this photo to 16 bits/channel first I still see the banding. I can only suppress the banding by filling the sky with a light gray color (for example RGB=140) plus a small amount of Gaussean noise (5%). After that the gradient is added.
It see no difference when I convert the image to 16 bit beforehand, I need the noise to prevent banding.

 

[138S]

When I convert this photo to 16 bits/channel first I still see the banding. I can only suppress the banding...

Just prevent banding by scanning/editing 16 bits, with levels placed properly.

 

[Billy Axeman]

Just prevent banding by scanning/editing 16 bits, with levels placed properly.

I can simulate that by creating a new image in my editor which has 16 bits/channel. When I add a gradient to it I see banding, I can only prevent that by adding some noise.

p.s. When I first started exploring this (a few years ago) I vaguely remember converting the photo from 8 to 16 bit prevented banding mostly, but I can't reproduce that. Perhaps there is also a factor which is editor dependent. Or my simulation by adding a gradient artificially has a flaw.

 

[138S]

Perhaps there is also a factor which is editor dependent.

Advanced editors (Photoshop, GIMP, Photopaint) allow non destructive edition.

If you adjust several times the curves in the regular way you perform destructive edition, for example if you compress too much shadows and later you expand then you may lose detail...

Instead you may perform a "non destructive" edition, if you place the curve edition in an adjustment layer then each time you modify the curve you depart again from the original image, so the repetitive editions don't provocate a cumulative degradation.

A Pro edition of important images, say a proficient edition of an image requiring multiple refinements/operations, should be done in the non-destructive edition way, the counter may deliver a botched job not in line of what it would be a good practice.

I would recommend to use Ps layers for edition of important images, not only you don't degradate the image quality, it also allows an amazing flexibiltiy and productivity when refining the image.

Let me show an example... you edit the shadows in an additional layer, later you edit the highlights in another layer taking the result from the first two layers. Then adjust the mix of the original layer, the shadow edition layer and the highlight edition layer to adjust a good balance of the shadows vs highlights. You may refine the final result in an easy and non-destructive way. If not doing that you would accumulate many curve editions and cumulative usage of the local exposure tools, and each time you go forward and backward to find the good point you destroy the image...

A good investment is spending around 8 hours in learning layers, etc with a good book or tutorial.

 

[Billy Axeman]

@alanrockwood, after a night of sleep I'm starting to see what you explained about banding in your previous posts. I did some supplementary simulations, with 8 and 16 bit images, conversion from 8 to 16 bit, and levels of noise and now I get logical results that are reproducible. So, my initial posts #41,43 were a bit naive and I appreciate your patience to explain it all to us noobs.

 

[Alan Edward Klein]

@alanrockwood, after a night of sleep I'm starting to see what you explained about banding in your previous posts. I did some supplementary simulations, with 8 and 16 bit images, conversion from 8 to 16 bit, and levels of noise and now I get logical results that are reproducible. So, my initial posts #41,43 were a bit naive and I appreciate your patience to explain it all to us noobs.

So what are your conclusions?

 

[Kodachromeguy]

I know that some won't believe what I am saying. If anyone would like to prove me wrong and would like to do a little experiment, here's what you can do.

Find or create a negative of an object with a smooth gradient. Scan the negative twice with a high resolution film scanner, like a 4000 dpi Nikon or Canon film scanner. For one scan do it in 8 bit mode, saving as a tiff. For the other scan do it in 16 bit mode and save it as a tiff.

Read the two negatives into photoshop, or whatever your favorite image processing application is. Before you do anything else, convert the 8 bit image to 16 bits. From now on keep that image in 16 bit format, including any saves. From now on you will be treating the two images in exactly the same way.

Now do some image manipulation on the two images, the one that was converted to 16 bits and the one that was created in 16 bits. Do the manipulation using the histogram correction method. That way you can do the exact same manipulations to the two images.

Are you able to produce banding in one but not the other image? Can you even see a difference between the two manipulated images by visual inspection?

Try it with Tri-X film, then FP-4 plus, then Tmax 100.

My prediction is that you won't be able to produce banding with Tri-X or FP-4 plus, and probably not with Tmax 100.

Something like Velvia might give a different result, but Velvia isn't a negative film.

Now for a related thought. Have any of you ever scanned a negative (any negative) using a high dpi scanner in 8 bit mode, and then converted it to 16 bit mode, keeping it in 16 bit mode from then on, and then been able to produce banding by manipulating the image?

This is very interesting. But this represents quite a bit of work. I don't understand why this is even an issue? Why not always scan 16 bit? Storage nowadays is cheap, so just scan at 16 and save the files as uncompressed TIFF. Am I missing something?

 

[Billy Axeman]

So what are your conclusions?

When things are kept simple and practical there are two main variables which control banding: using 8 or 16 bit/channel, and the level of noise in the image.

The need for selecting 8 bit or 16 bit, or adding supplementary noise to avoid banding depends on the initial level of noise or grain in the image. This level of noise varies with the source of the image: after scanning (film), or the film format, or directly from the camera (digital), or a digital image photographed at low or very high ISO's etc.

An 8 bit image needs more noise or grain to be present to prevent banding than a 16 bit image. But it is possible that 8 bit is already sufficient when the image has a lot of noise. So, you don't always need a 16 bit image in this respect (there can be other reasons to use 16 bit though).

An 8 bit image can be converted to 16 bit afterwards to help prevent banding, but again it depends on the level of the existing noise or grain if you need to add extra noise. So for example a digital image directly from the camera, no matter 8 or 16 bit, almost always needs noise to be added because it is virtually noise free compared to film (unless it is shot at a very high ISO value).

To get a feeling for this banding phenomenon, it is possible to completely simulate this in an editor by making a new image (8 or 16 bit), filling it with a light gray value, adding (Gaussian) noise at a certain level, and adding a gradient in the end to check banding. All possible variations with 8 or 16 bit images, conversion from 8 to 16 bit afterwards, and noise level, can be tested quickly to see the result without bothering about the source (film type, digital) and/or the scanning method.

Of course you need a separate test for a specific source to see the effect. But I think it is advantageous to always work with 16 bit images because you need less noise to prevent banding. Ultimately adding noise is not what you want to do with a photo. However, it is also possible to prepare only part of the image, for example when you want to replace a blown out sky with a gradient.

All of this is based on only a few tests after my interpretation of what @alanrockwood has said, so bear with me when I miss a point.

 

[Alan Edward Klein]

Thanks. I guess that means no conclusion.

 

[138S]

This is very interesting. But this represents quite a bit of work. I don't understand why this is even an issue? Why not always scan 16 bit? Storage nowadays is cheap, so just scan at 16 and save the files as uncompressed TIFF. Am I missing something?

Thanks. I guess that means no conclusion.

There is a conclusion: scan 16 bits, edit a 16 bits/ch oversampled image, practice non destructive edition, learn how to perform smart sharpenings... and we have the basic tools for a top Pro workflow. That was the easy part.

What is difficult is having a sound aesthetic criterion to obtain a great intepretation, many times it's also about having respect for the scene, for the subject or for the light that made the image. We can enhance the image a lot without being intrusive, or sometimes we have good reasons to be intrusive and that intrusion itself is art.

Sometimes I edit the images for a certain artist and I always get amazed about how he has clear ideas about what has to be done with the image, it's about pre visualizing the final result and realizing the emotions it has to comunicate before you start with the adjustments. I learned from these experiences, but I realize that one has to be a bit gifted to shine in that.