An alternative to Negative Lab Pro and Lr has to exist (C-41 reversal and orange mask removal)?!

[Photo Engineer]

Exactly. I’ve written code professionally for a very long time and this ain’t my first software rodeo. I do in fact have really long, really hard experience in dealing with the software world and end users in general. Lots of guys probably would be willing to pay and not post a copy of it online, but it only takes 1 guy to do that, and then it’s over, unless you’ve spent even more time and money locking it down with license keys.

Also, it’s not like you can’t get the results from my software. I don’t charge an arm and a leg to dev and scan, or if you’ve already got developed film, to just scan it. In some ways that’s my software subscription model.

And, as I learned, software patents are very difficult to obtain!

PE

 

[Helge]

And even those aren't very good compared to what can be done manually. The difference is quite startling in both quality & accuracy of colour.

No, Flextights are quite overrated on the whole. They are clearly inferior to drum scanners, which again are a very mixed (and expensive) lot.

The current microscope/macro thing really is promising, and you can get unbelievable results with humble equipment and a bit of perseverance and knowledge.

A scanner involving an un-bayered B&W sensor, need not be very large at all, will probably be the future of easy, inexpensive and accessible scanning.
The sensor might either be brought into direct contact with the emulsion, liquid immersed, or, a small but sharp macro lens, also preferably liquid immersed could be employed.
Both walking in steps across the frame, with overlap to aid stitching and alignment.
Any kind of light source might be used in the small spot under the target and sensor to get the effect wanted or data extracted.
Wideband or narrowband RGB, IR, UV, collimated, diffuse etc.

 

[Adrian Bacon]

how well does the output resemble an optical RA-4 print (aka, removing the mask as a mask, not an overall colour cast - I take it that's what your grey card linearisation is aimed at?);

For the most part it looks like a picture taken with a digital camera that was shot raw and run through Adobe Camera Raw. The mask effectively is a color cast. There are two things at play here that trip most people up, the orange mask and the gamma of each channel. The gamma of each channel is rarely the same. They're usually relatively close, but not the same. If you apply the multipliers and then just invert it, you'll see that the blacks and shadows are actually correct and as you add more density, the colors start to diverge and color casts start to show up. If you adjust the gamma of each channel so that they're the same then apply the multipliers the color casts go away, assuming you shot the exposure cards at the same white balance as the film. From there, it's just a matter of making the gamma of each channel linear relative to the exposure that made it. If this is done correctly, in floating point linear light space, a correctly exposed (via incident reading) 18% grey card is 0.18, and each stop of exposure up or down doubles or halves that. My code does all this in one operation via a look up table that takes the raw integer samples as the index into the table and outputs the floating point values. The look up table is a basic text file so I can change the gamma of each channel and basically do whatever I want without actually changing any of the code. This is literally a handful of lines of code to suck in the lookup table and turn it into an in-memory array with the values and a handful of lines of code to step through each pixel and get the floating point value out of the array. The biggest challenge is figuring out the gamma of each channel so you can make the lookup table. That part you have to do for each emulsion you want to handle.

I'm not sure whether the colour checker is the correct choice for correctly representing the different fundamental colour balances of different films - the whole point of the colour checker profiling system is (as far as I understand it) to attempt to eliminate differences in colour reproduction between emulsions/ sensors - how well does it do this within your setup?

The color checker is so you can look at the red/green/blue/yellow/magenta/cyan patches and adjust the hue and saturation to your liking. Depending on what color space you picked as your source model, they'll either be a little off, or a lot off. If you want to do it via a 3D lut, you can, but it's just simpler to have a basic lookup table for the hues, and one for the saturation. Each film will have its variation of colors depending on the emulsion. You can use the hue and saturation look up tables to adjust it. Again, it's a handful of lines of code to suck in the file and turn it into an array, then another handful of lines of code to step through each pixel, calculate its current hue angle, use that as an index into the array to get how many degrees to adjust it, then adjust it, and do the same thing for saturation adjustment. The hue and saturation calculation algorithms are available via google as are the hue and saturation adjustment algorithms. If you google hard enough you don't even have to write that code as there are a multitude of examples of how to do it already written in C/C++ and pretty much any other language you would care to do this in. In my particular case, I have a set of C-41 generic tables that I use as a base and then modify from there as needed, though usually I just let it be as it's close enough for most emulsions. Again, I generally supply the ACR look in broad strokes, but in reality, you can make it look any way you want. I arrived at the generic lookup tables by actually starting to profile each emulsion to make an exact ACR look. After a few emulsions, it became pretty clear that there was a pattern of similar changes, so I took the ones that I had done and averaged them to make the generic. Each emulsion will vary a little from this generic, so each film does look different from each other using the same generic profile, but if you want to, you can totally dial that out or exaggerate it by making a set of tables specifically for that emulsion. In a way, that generic c-41 is basically similar RA-4 paper. Each emulsion for a given paper will vary how it looks a little because the paper and emulsion aren't exact mirrors to each other. My route is a little (a lot) different, but has the same effect. As a side note, what I'm doing here is actually exactly how LR and the DNG spec supplies it's various looks to raw files. The only difference is that I'm doing it in raw unmanaged space to conform things to a color space, and LR is doing it within the context of its internal color space to provide a look on top of it's flat ACR look.

 

[Adrian Bacon]

And, as I learned, software patents are very difficult to obtain!

PE

yep.

 

[Lachlan Young]

No, Flextights are quite overrated on the whole. They are clearly inferior to drum scanners, which again are a very mixed (and expensive) lot.

The current microscope/macro thing really is promising, and you can get unbelievable results with humble equipment and a bit of perseverance and knowledge.

A scanner involving an un-bayered B&W sensor, need not be very large at all, will probably be the future of easy, inexpensive and accessible scanning.
The sensor might either be brought into direct contact with the emulsion, liquid immersed, or, a small but sharp macro lens, also preferably liquid immersed could be employed.
Both walking in steps across the frame, with overlap to aid stitching and alignment.
Any kind of light source might be used in the small spot under the target and sensor to get the effect wanted or data extracted.
Wideband or narrowband RGB, IR, UV, collimated, diffuse etc.

The Flextight is actually very, very good relative to many high end CCD & drum scanners, but a lot is dependent on actively preventing Flexcolor doing things to your file before you get it into Photoshop etc & having a competent operator. The shortcomings mainly relate to the resolution limitations (though you can work round that) and the slightly cleaner shadows from PMT's. The inversions done in Flexcolor have colour issues & are significantly noisier than if you get an un-inverted tiff into Photoshop & work from there. You can even get to deliver files without embedded colour profiles etc.

 

[Helge]

The Flextight is actually very, very good relative to many high end CCD & drum scanners, but a lot is dependent on actively preventing Flexcolor doing things to your file before you get it into Photoshop etc & having a competent operator. The shortcomings mainly relate to the resolution limitations (though you can work round that) and the slightly cleaner shadows from PMT's. The inversions done in Flexcolor have colour issues & are significantly noisier than if you get an un-inverted tiff into Photoshop & work from there. You can even get to deliver files without embedded colour profiles etc.

Certainly they are better than most consumer scanners, but at 10X or 20X the price they better damn well be.
Drum scanners are a crapshoot. They might be fantastic due to the still unmatched performance of photo multiplier tubes. Or they might deliver average results at best, if not maintained well or operated correctly.
Resolution is where macro scanning beats Flextights and a lot of drum scanners hands down. That's before we talk about the massive cost difference.

 

[Adrian Bacon]

Certainly they are better than most consumer scanners, but at 10X or 20X the price they better damn well be.
Drum scanners are a crapshoot. They might be fantastic due to the still unmatched performance of photo multiplier tubes. Or they might deliver average results at best, if not maintained well or operated correctly.
Resolution is where macro scanning beats Flextights and a lot of drum scanners hands down. That's before we talk about the massive cost difference.

Not to mention speed of scanning difference

 

[Lachlan Young]

The mask effectively is a color cast.

Yes - but it's inversely proportional to exposure, so it cannot be removed as a simple global colour correction - which an awful lot of the proprietary software on the market fails to understand.

Resolution is where macro scanning beats Flextights and a lot of drum scanners hands down. That's before we talk about the massive cost difference

Yes, no & maybe - speed & upgradeability are definite positives for macro scanning - & the very low noise floor of currently CMOS sensors too. The problem is the Bayer array & how to get 4:4:4 type of output so that you get a fighting chance of competing with 3x individually filtered CCD or PMT in colour depth & representation without having to severely downsample a single shot or buy a BW Phase One & separation filters - multisampling with the sensors that can do xy stitching using their vibration control is probably the answer - which, ironically enough is essentially how a Frontier's sensor works.

 

[Adrian Bacon]

Yes - but it's inversely proportional to exposure, so it cannot be removed as a simple global colour correction - which an awful lot of the proprietary software on the market fails to understand.

Maybe I’m just a dummy and fall in the fail to correctly understand pool, but what I previously described seems to deal with it just fine, or at least well enough that myself and a whole lotta other people aren’t complaining about the result.

If you’d care to elaborate on how a global color correction doesn’t do the job, I’d probably find it pretty instructive.

 

[Patrick Robert James]

You can do everything you want with Raw Therapee if you want free. It is pretty complicated though so if you don't understand what is going on you will probably get lost. It was the best for the trials I did. I am not much for making camera scans though since I rarely shoot color neg, and that is almost always 35mm. The Nikon scanner I have is good enough. Camera scans though extract a whole other level of quality if you want to take the time to do them right. If I had a 50 megapickle camera i'd probably do them, but that would still require a bunch of stitching. The Bayer array causes issues as well but i am not going to get into it here.

NLP just seems to me to be from someone who can code but doesn't understand color. Everything I've seen converted from it has terrible color. There is something flawed with the program. Just my opinion.

It is too bad Adrian doesn't make his program available, but I understand. Who wants the grief? Just imagine all the yahoos causing problems.

 

[Helge]

Yes, no & maybe - speed & upgradeability are definite positives for macro scanning - & the very low noise floor of currently CMOS sensors too. The problem is the Bayer array & how to get 4:4:4 type of output so that you get a fighting chance of competing with 3x individually filtered CCD or PMT in colour depth & representation without having to severely downsample a single shot or buy a BW Phase One & separation filters - multisampling with the sensors that can do xy stitching using their vibration control is probably the answer - which, ironically enough is essentially how a Frontier's sensor works.

That is absolutely a concern and is why I, as mentioned in a previous post, suspect that commercial implementations of the idea will use a small “phone” CMOS without bayer filter, but with variable backlight.
As for now the solution is to either have such high magnifications that the Bayer filter has minimal effect.
Or alternatively (I’ve not tried this though), shift and rotate the frame slightly multible times per shot over the same part with a subsequent merge, to alleviate the bayer effects and pixel array aliasing too.

Some people have used sensor shift to get higher resolution and not have bayer effects. I’ve not been entirely convinced by the results though.

 

[Adrian Bacon]

That is absolutely a concern and is why I, as mentioned in a previous post, suspect that commercial implementations of the idea will use a small “phone” CMOS without bayer filter, but with variable backlight.
As for now the solution is to either have such high magnifications that the Bayer filter has minimal effect. Or alternatively (I’ve not tried this though), shift and rotate the frame slightly multible times per shot over the same part, to alleviate the bayer effects and pixel array aliasing too.

There seems to be a fair amount of concern over using a bayer array and it causing problems with scanning film. As a user of a bayer array based scanning system, it's much less of an issue than what people probably think it is. A modern APS-C based camera has been at the 24MP mark for quite some time. At 240 pixels per inch print resolution, it will make a very nice looking 16x24 inch print of 35mm or 16x20 print of a 6x7 120 frame. Modern Bayer interpolation algorithms are very, very good. While 16x20 is not a large print by large print standards, it most definitely is not a small print, and based on the number of requests I get to make prints, is absolutely on the large end of what is generally printed. 5x7 followed very closely by 8x10 and 11x14 by a distant third is what I get most requested to print. Every once in a while I'll get a request to do something larger, but I easily print hundreds of the other sizes for every 11x14, and dozens of 11x14s for every larger request.

Don't get me wrong, I'm generally all for more resolution, but there seems to be a bit of a disconnect between how much resolution people want to scan at and how much resolution you actually need for common stuff. If you're worried about the bayer array causing weird aliasing artifacts with the film grain, you maybe might have to worry about that if the particular camera you're using doesn't have a low pass filter (like some Nikon models), but for pretty much everything else, there's a low pass filter that does a very good job mitigating problems without completely destroying fine detail. The camera manufacturers are not dummies when it comes to this sort of thing. Combine that with the somewhat random nature of the grains/dye clouds and the modern AHD Bayer interpolation and the whole "artifacts from the bayer array" thing hasn't really been much of an issue in my experience. Of course I get the occasional comment from a customer about how big the grain of some emulsion is when somebody takes one of my DNGs and looks at it at 1:1 or 2:1 in the LR develop module. I inevitably have to gently remind them that when they do that, they are effectively blowing it up to a giant print (by giant print standards) and looking at that with a magnifying glass, and that maybe a more realistic way to look at it is to put it full screen on their big 60-70 inch 4K TV and see what that looks like instead. Almost always, once they do that they comment that it looks a lot better that way and is pretty sharp. And that's with a 35mm frame size. A 4K TV is ~8MP and looks really good, even at the large end of the TV scale.

The point is, more resolution is generally better, but in reality, for most display purposes, you actually need a lot less than you think you do, and 24MP covers a really large swath of that. If you want to scan 120 and 4x5 with a 50MP Canon 5Ds, you can and it will look better than a 24MP scan at really large print sizes, but again, now we're getting into truly ginormous prints. Just some food for thought.

 

[Carriage]

Maybe I’m just a dummy and fall in the fail to correctly understand pool, but what I previously described seems to deal with it just fine, or at least well enough that myself and a whole lotta other people aren’t complaining about the result.

If you’d care to elaborate on how a global color correction doesn’t do the job, I’d probably find it pretty instructive.

I'd be interested too. When printing in the darkroom I can't see how compensation for the mask isn't global so I don't know why doing that digitally isn't going to work.

 

[Lachlan Young]

I'd be interested too. When printing in the darkroom I can't see how compensation for the mask isn't global so I don't know why doing that digitally isn't going to work.

It's because the mask exists to correct imperfections in coupler formation in shadow values, not highlights - when printed to chromogenic papers. Hence a global correction works in the analogue darkroom because the mask is intended to colour correct the shadows to run in line with the highlights on chromogenic paper. Not the case with scanning. Without the coupler interactions, you will get off colour at one or other end of the scale if you correct for an overall colour cast. Dividing out or otherwise proportionally correcting for the mask becomes necessary. Once that's done, the astonishing quality of colour representation that C-41 can deliver is unleashed.

 

[Photo Engineer]

I'm sorry, but color correction in negative films is global and creates a positive correction curve that matches the negative curve point by point. I see no error or difference of any substantial note when I print negatives digitally and compare them to darkroom prints.

PE

 

[MattKing]

Correct me if I am wrong, but I think what is being said here is that the mask isn't a single colour overlaid on the image, but rather a correction curve that applies more colour correction to some parts of the image then at others.
As I understand it, the characteristics of that curve come from the interaction between the characteristics of the film emulsion and the characteristics of the RA4 paper emulsion.

 

[Lachlan Young]

Correct me if I am wrong, but I think what is being said here is that the mask isn't a single colour overlaid on the image, but rather a correction curve that applies more colour correction to some parts of the image then at others.

Yes - or at least that's how it seems to behave in my experience.

 

[Photo Engineer]

The RA4 emulsions are indifferent to the mask. They can print unmasked film or masked film and get good results but the unmasked results will have poorer color. I have run those tests with old unmasked Sakura and Agfa films.

PE

 

[Adrian Bacon]

I'd be interested too. When printing in the darkroom I can't see how compensation for the mask isn't global so I don't know why doing that digitally isn't going to work.

That's what I'm thinking too. When you make an RA-4 print and you're using the color filters, you're making global color corrections (albeit analog ones). How is that any different than doing raw global color corrections digitally? Even looking at Kodaks characteristic curves in their tech sheets, if you apply a multiplier so that each channels floor (i.e. the mask) matches, the rest of the curves line up with each other with the exception of the gamma differences of each color.

Am I missing something here?

 

[MattKing]

That's what I'm thinking too. When you make an RA-4 print and you're using the color filters, you're making global color corrections (albeit analog ones). How is that any different than doing raw global color corrections digitally? Even looking at Kodaks characteristic curves in their tech sheets, if you apply a multiplier so that each channels floor (i.e. the mask) matches, the rest of the curves line up with each other with the exception of the gamma differences of each color.

Am I missing something here?

Again correct me if I am wrong...
The pure optical route is:
1) A C41 film, with its built in colour response curve;
2) A colour correcting mask, with a built in colour response curve;
3) Colour printing filters, which to all intents and purposes apply a linear correction; and
4) RA4 paper, with a built in colour response curve.
None of those above three curves are linear.
If you proceed using a film to digitization procedure, there are curves inherent in the film and the mask, but the other steps start out as linear.

 

[GLS]

Resolution is where macro scanning beats Flextights and a lot of drum scanners hands down. That's before we talk about the massive cost difference.

Yes, I've been very satisfied with my DSLR/macro digitisation method, and that's even without doing multiple shots + stitching.

 

[Photo Engineer]

Electronic imaging (scanning or cameras for example) follow a V LogE curve where V = Voltage. This results in a curve much like a D LogE curve. We plotted them at EK. You can get them from some scanners I am told. High end ones of course.

PE

 

[Adrian Bacon]

Again correct me if I am wrong...
The pure optical route is:
1) A C41 film, with its built in colour response curve;
2) A colour correcting mask, with a built in colour response curve;
3) Colour printing filters, which to all intents and purposes apply a linear correction; and
4) RA4 paper, with a built in colour response curve.
None of those above three curves are linear.
If you proceed using a film to digitization procedure, there are curves inherent in the film and the mask, but the other steps start out as linear.

I’m not entirely sure 2. Is completely correct. If the mask has a response curve as opposed to a global multiplier offset, wouldn’t the paper need the opposite of the curve in addition to its inherent color response curve (IE a mask)? If that’s the case, then why the color filters? They behave as a global multiplier and wouldn’t compensate for any mask curve, so if there is a mask curve where is it being dialed out in optical land?

 

[Photo Engineer]

Paper does NOT need the opposite of the curve. All it needs is a speed offset in the B and G layers to account for the mask. Thus these layers are faster to compensate for the basic orange cast of the negative's mask. This is also why you can print from unmasked negatives and only suffer from the inherent impurities of the dyes.

PE

 

[Adrian Bacon]

Paper does NOT need the opposite of the curve. All it needs is a speed offset in the B and G layers to account for the mask. Thus these layers are faster to compensate for the basic orange cast of the negative's mask. This is also why you can print from unmasked negatives and only suffer from the inherent impurities of the dyes.

PE

I’m interpreting “speed offset in the layers” as a global multiplier if done digitally, as the net result is the orange cast is effectively dialed out as a result of the multiplication that is effectively happening as a result of the layer speed differences.

If that is the case, then doing a raw global multiplier on each channel to dial the mask out isn’t totally wrong, as that produces the same effect as having each layer a different speed.

The mask itself still might have a response curve, but that doesn’t appear to be doing any harm to the image, at least in my setup.

Feel free to correct, as I’m finding this discussion to be very interesting