Digital Negatives: Clipping when applying curves

[aconbere]

Looking for some advice on working with digital negatives. (Note that this is for a class on building curves for processes, we're using Ilford MG RC Deluxe as an easy paper to work with). So what's happening here is I've built up a curve profile to linearize contrast for this paper. I took an image I wanted to print, loaded it in photoshop, applied the curve I'd built, and it ends up clipping my highlights in ugly ways. Annnnd I'm just not familiar enough with photoshop to know if it's like there's something I've done wrong, or how to get myself into a better place.

As an example here is a crop of a highlight in the original digital image.

Then when I apply the curve. It's not super easy to see, but the clipping is there in the center as speckling.

Finally the print which accentuates the speckling.

Anyway, just hoping someone has some advice for me

 

[koraks]

Are you working in 16 bit bit depth? If you're working in 8 bit, it's common to run into problems with posterization, which this essentially is. It's best to start with a 16 bit depth image; if all you have is an 8 bit image, then ensure the first step you take is to scale it up to 16 bit before applying any adjustment curves.

Part of the problem is also that the curve you've applied is kind of odd; it flattens out the upper midtones very dramatically. Maybe that's an intended effect, IDK, but it does exacerbate the postrization problem since flattening out one bit of the curve means that another bit needs to have a very steep gradient.

Note flattened out bit marked in red and steep bit in blue.

 

[aconbere]

Are you working in 16 bit bit depth? If you're working in 8 bit, it's common to run into problems with posterization, which this essentially is. It's best to start with a 16 bit depth image; if all you have is an 8 bit image, then ensure the first step you take is to scale it up to 16 bit before applying any adjustment curves.

Part of the problem is also that the curve you've applied is kind of odd; it flattens out the upper midtones very dramatically. Maybe that's an intended effect, IDK, but it does exacerbate the postrization problem since flattening out one bit of the curve means that another bit needs to have a very steep gradient.
View attachment 383834
Note flattened out bit marked in red and steep bit in blue.

it’s funny you suggest 8bit depth because that’s exactly what I told my friend it looked like! The image starts as a camera raw so plenty of bit depth to work with there, but Im used to working in lightroom, maybe i screwed up the import to photoshop.

I’ll start there.

 

[Alan9940]

If this area is truly clipped, that is 255-255-255 then there is no adjustment you can make to get it back. Is it possible you could clone similar areas over the clipped area to "rebuild" it?

 

[aconbere]

If this area is truly clipped, that is 255-255-255 then there is no adjustment you can make to get it back. Is it possible you could clone similar areas over the clipped area to "rebuild" it?

It's truly clipping, but only /after/ the application of the curve. So the source image has sufficient information to produce a pleasant tonal curve, just when adjusted for printing in the process it's pushing the highlights to clip.

For now what I've done is just pulled the exposure down on the whole image a smidge which has prevented the worst of the issues (which are kind of ugly in this light, but area TERRIBLE anywhere on the subjects face). Maybe the answer longer term is to check for clipping especially in the highlights and selectively pull them down a bit.

 

[aconbere]

Are you working in 16 bit bit depth? If you're working in 8 bit, it's common to run into problems with posterization, which this essentially is. It's best to start with a 16 bit depth image; if all you have is an 8 bit image, then ensure the first step you take is to scale it up to 16 bit before applying any adjustment curves.

Part of the problem is also that the curve you've applied is kind of odd; it flattens out the upper midtones very dramatically. Maybe that's an intended effect, IDK, but it does exacerbate the postrization problem since flattening out one bit of the curve means that another bit needs to have a very steep gradient.
View attachment 383834
Note flattened out bit marked in red and steep bit in blue.

I checked and photoshop had converted the raw correctly into 16bit greyscale so bit depth shouldn't be the issue. To address your point about the flatspot on the curve, the image you're looking at is actually the source image before application of the curve, it's just a low contrast set of tones (It's a light in the background, it's going to be mostly zone 9-10). But I think you're getting at the right idea. Because that section of the curve for this paper is much flatter than normal, so we're taking something with only a small amount of tonal separation and then flattening that even further.

To be clear, I'm actually totally okay if hot point goes to paper white. But what happens in /practice/ is that there is no smooth gradient out of that clipped section and instead I end up with these blocky, pixelated chunks. (Very reminiscent of how low bit depth video ends up with chunky gradients in the blacks).

 

[koraks]

Good that you checked it. The story behind the 16 bit may be more complex since the nominal bit depth of the RAW file may be 16 bits, but this doesn't mean that it actually contains 16 bits of real data. It's very well possible (even likely) that the actual resolution is limited to e.g. 14 bits, with the remaining bits being padded. This means that posterization can set in a little earlier than you might expect. This could induce problems when trying to work with very extreme curve shapes (very flat and/or very steep - especially the latter).

Another issue that may emerge is interference patterns when downsizing or downsampling to a lower resolution, e.g. when printing at a relatively small size. This can result in jagged edges here and there, especially if there's an already existing pattern (like pronounced digital noise or film grain) in the image data. But I'd expect that things would render a little more smoothly - and it's of course moot if you already see the problem emerge as soon as you apply the compensation curve, before any further downscaling happens.

All considered I'm not entirely sure where the problem is, but it does look much like a simple case of trying to push the limits a little further than the image data allow for.

 

[aconbere]

Good that you checked it. The story behind the 16 bit may be more complex since the nominal bit depth of the RAW file may be 16 bits, but this doesn't mean that it actually contains 16 bits of real data. It's very well possible (even likely) that the actual resolution is limited to e.g. 14 bits, with the remaining bits being padded. This means that posterization can set in a little earlier than you might expect. This could induce problems when trying to work with very extreme curve shapes (very flat and/or very steep - especially the latter).

Another issue that may emerge is interference patterns when downsizing or downsampling to a lower resolution, e.g. when printing at a relatively small size. This can result in jagged edges here and there, especially if there's an already existing pattern (like pronounced digital noise or film grain) in the image data. But I'd expect that things would render a little more smoothly - and it's of course moot if you already see the problem emerge as soon as you apply the compensation curve, before any further downscaling happens.

All considered I'm not entirely sure where the problem is, but it does look much like a simple case of trying to push the limits a little further than the image data allow for.

So yeah, I was curious because it gets /much/ worse in the contact print, I feel like it's possible that down sampling to 300dpi could be making it worse. Is it worth it to try printing the digital negative at a higher dpi?

I am also curious if you think it would be worth it for me to try a proper flatbed scan. In this case this is a raw produced from a camera scan on a Fuji XT-3 of a 4x5 negative, which is way more than adequate for what I generally use scans for (proofs and sharing with friends). But I could make a proper flat bed scan of this negative at whatever the ungodly resolution of that is. In that case I should get a proper 16bit tiff out of it, and a lot more resolution to work with for better tonal clarity.

 

[koraks]

Is it worth it to try printing the digital negative at a higher dpi?

Yeah, give it a try; it might make a bit of a difference in this particular case. BTW, when outputting negatives to my Epson printer, I either try to fix the resolution to its native 1440dpi (or 2880, although this generally doesn't yield much more usable detail), or I just send the file to the printer in whatever resolution it happens to be. I use the former for self-generated images like step tablets and test images, the latter for real photos.

As to the scan: ah, yes, I see - so this is basically a digital photo of a negative. Then it's virtually certain to be far, far less than 16 bits per pixel effective data. Most negatives have a fairly limited dynamic range (less than 2.0logD or around 7 stops, usually a bit less), while the digital camera is optimized to capture a fairly large dynamic range (10+ stops). So best case you're working with maybe 12-13 bits of actual resolution instead of the 16 bits you may be expecting. It could be worthwhile trying to scan the negative with a flatbed scanner and during scanning, ensure to scan at 16 bit/px and to set the black and white points so that the image data occupy the entire space. Coincidentally, I just wrote something about this here (although in a different context): https://www.photrio.com/forum/threads/doing-color-negative-inversions-manually.210239/post-2846671 - but feel free to skip that bit of theoretical musings and just give it a try.

 

[aconbere]

Yeah, give it a try; it might make a bit of a difference in this particular case. BTW, when outputting negatives to my Epson printer, I either try to fix the resolution to its native 1440dpi (or 2880, although this generally doesn't yield much more usable detail), or I just send the file to the printer in whatever resolution it happens to be. I use the former for self-generated images like step tablets and test images, the latter for real photos.

As to the scan: ah, yes, I see - so this is basically a digital photo of a negative. Then it's virtually certain to be far, far less than 16 bits per pixel effective data. Most negatives have a fairly limited dynamic range (less than 2.0logD or around 7 stops, usually a bit less), while the digital camera is optimized to capture a fairly large dynamic range (10+ stops). So best case you're working with maybe 12-13 bits of actual resolution instead of the 16 bits you may be expecting. It could be worthwhile trying to scan the negative with a flatbed scanner and during scanning, ensure to scan at 16 bit/px and to set the black and white points so that the image data occupy the entire space. Coincidentally, I just wrote something about this here (although in a different context): https://www.photrio.com/forum/threads/doing-color-negative-inversions-manually.210239/post-2846671 - but feel free to skip that bit of theoretical musings and just give it a try.

I'll give it a shot! Can't hurt certainly. Really appreciate the help.

 

[koraks]

Sure thing; please report back if it helps any. It would be useful to know.

Btw, I've run into similar things when very dramatic adjustments were required by certain printing processes, especially if they were different to control. The transition from tone to paper white on pigment processes like carbon is notorious and easily gets me into this pickle. You sometimes really have to work hard to get a decent compromise between which ink channels the printer uses to lay down density, target printing process parameters and the curve adjustments.

 

[Alan9940]

It's truly clipping, but only /after/ the application of the curve. So the source image has sufficient information to produce a pleasant tonal curve, just when adjusted for printing in the process it's pushing the highlights to clip.

Ah, I missed the fact that it was only after applying a curve. Sounds like you might have worked something out already, but you might want to try pulling down slightly on the highlight node on the upper right. Keep the Info panel open and verify the tones with a 3x3 eyedropper.

 

[Carnie Bob]

I hesitate to add my 2 cents here but here goes..

It kind of sounds like the OP is double profiling which can always end in disaster.
If he or she has a good original profile that is calibrated, then a second curve makes no sense to me... for all my work whether it is pt pd, gum, cyanotype , silver we use the same calibrated profile that was produced by creating a linearized 100step wedge where the input and output correspond within limitations of the process. LAB 96- LAB 100 will always clip no matter what so all highlight with detail is kept under LAB 95 for safety. I have printed thousands of images with different end process with the same custom profile, originally made for me here at my shop by Ron Reeder and in 2020 I made a new one with Greg Brophy when I changed printers.
When making negatives from digital files the old adage garbage in garbage out applies, I see many photographers get away with monster corrections to files and printing onto matt inkjet paper but when those same images get separated out posterization, glitches start. I am not sure why but probably the stripping apart the image is too much for it to handle.

So here we make the file look great in PS and then when making a neg via Print Tool we apply the appropriate bit order curve, Only once.
Koraks in post 11 kind of confirms my suspicion.

 

[aconbere]

I hesitate to add my 2 cents here but here goes..

It kind of sounds like the OP is double profiling which can always end in disaster.
If he or she has a good original profile that is calibrated, then a second curve makes no sense to me... for all my work whether it is pt pd, gum, cyanotype , silver we use the same calibrated profile that was produced by creating a linearized 100step wedge where the input and output correspond within limitations of the process. LAB 96- LAB 100 will always clip no matter what so all highlight with detail is kept under LAB 95 for safety. I have printed thousands of images with different end process with the same custom profile, originally made for me here at my shop by Ron Reeder and in 2020 I made a new one with Greg Brophy when I changed printers.
When making negatives from digital files the old adage garbage in garbage out applies, I see many photographers get away with monster corrections to files and printing onto matt inkjet paper but when those same images get separated out posterization, glitches start. I am not sure why but probably the stripping apart the image is too much for it to handle.

So here we make the file look great in PS and then when making a neg via Print Tool we apply the appropriate bit order curve, Only once.
Koraks in post 11 kind of confirms my suspicion.

What’s double profiling? I’m super new to this (and also not very experienced with photoshop) so there is a very high likelihood I’m messing something up along the way.

The curve I’m using I’ve built myself by iterating on 100step wedges and could also have issues!

I did go make some nicer scans yesterday and i’ll post a similar crop of that light later today.

 

[koraks]

What’s double profiling?

I think he refers to the situation where you (possibly inadvertently) apply two separate adjustment curves on top of each other. However, that doesn't appear to be the problem you're running into, as it would only become apparent after the second curve is applied, for instance during printing (e.g. using QTR with Epson printers). You generally only notice it in the final print, but in your case, you see it happening immediately when doing the first (and possibly, only) adjustment.

 

[aconbere]

I think he refers to the situation where you (possibly inadvertently) apply two separate adjustment curves on top of each other. However, that doesn't appear to be the problem you're running into, as it would only become apparent after the second curve is applied, for instance during printing (e.g. using QTR with Epson printers). You generally only notice it in the final print, but in your case, you see it happening immediately when doing the first (and possibly, only) adjustment.

Got it. I am doing two adjustments, but the first is just the necessary adjustments to make a positive from the negative scan. Perhaps the point is that if I do a poor job of that conversion then when I apply my process curve it's going to highlight those failures.

Getting a better scan has helped:

You can still see where there is a smattering of pure white areas, but the tonal gradation between them and the surrounding area is much improved. Banding remain on the edges of the light so I guess I'll see how that turns out. It's also not a perfect comparison because along the way I'm getting smarter about watching for those highlights in my edits and protecting them with a little padding when making my original adjustments on the negative.

I should have a chance to do a new print soon and I'll try printing with downscaling to 300dpi, hopefully that helps with some of that posterization we saw. And once again I'll report back.

Thanks!

 

[koraks]

Yes, I can see how that may look better, although it's hard to tell without context and without being able to see what the final print might look like.
There's still substantial posterization going on, but it has now shifted to the concentric zone between the dark surrounding and the light heart. Whether this will be a problem in the final print, you'll have to figure out by printing the negative and seeing how it pans out.

 

[aconbere]

Okay! I have some small success to report.

I got to sit down and really spend some time reading and testing different settings on the scanner. First, I fixed a persistent focus issue with my scanner dialing in the film holder height. Second in previous attempts I hadn't been using the multiple exposure mode, and that seems to have much improved the shadow details (highlights when flipped).

There's still banding, but it's further reduced, and then reduced again by giving the curve a nudge to give the highlights a little more separation.

Anyway, I think this is the best I can do. Importantly this is the very worst of the whole image, and non important (I could just digitally remove it). But every improvement here is improving tonal range in important areas like... skin.

Will report back when I print this... probably monday.

- A

 

[koraks]

OK, that's some progress; at least now the posterization is hiding in a dithered pattern that makes it less obnoxious. It's still pretty bad and much more severe than I'd expect with a 12+ bit depth input. Something's not quite right with the digital end of this process, that's for sure.