on.raw.photography

AlexisMagni wrote: Thu Sep 04, 2025 7:48 am It seems everyone was waiting for me to show up with these questions!

Now we're waiting for you with the second question: digital noise

Edit C.O.: That is digital noise when using PerfectRAW on digital images, not ColorNeg DC etc.

I'm asking the question about digital noise.

This example is a photograph taken at ISO 6400, with a Fujifilm X-Pro2 camera and a Leica-M 35mm Summicron Asph. lens.
Digital noise at this sensitivity is inevitably visible on any camera.

The question is: how visible is it in a print at a nominal magnification?
Specifically, a print from a 24 megapixel camera at 300 ppi, which equates to a size of approximately 34 x 51 cm (13.39 x 20.10 inches). Or any other camera with any nominal size.

Example #2 is a 100% crop of the nominal print size, 34 x 51 cm (13.39 x 20.10 inches). Both luminance and chrominance noise are evident.
But from my point of view, in this print size, it is not dramatic.
But I'm not sure that the noise we see on a monitor is the same as what we would see in a photographic print. It's more likely that in a print, this noise is somewhat attenuated, or perhaps attenuated enough, to the point of seeing some grain, but not all of this luminance and chrominance noise.
I don't even think you can see only luminance grain, which is more similar to film grain.
But I also think that chrominance noise, that is, colored noise, is not as evident as on a monitor, thus making the print very acceptable, and perhaps even pleasant, like a high-sensitivity analog print.

This thesis is, from my point of view, and for how I understand photography, absolutely valid at lower sensitivities. Sensitivities are always higher than the camera's nominal sensitivity, but perhaps not as high as ISO 6400.
So, how much further can you push the sensitivity while still obtaining a pleasing, grainy print similar to high-sensitivity analog photography, without necessarily incurring intrusive and unpleasant noise?

Why would you print a noisy photo when today's software is capable of perfectly cleaning it up, even preserving or recreating detail?

If we're talking about art, not commercial photography, or if we're talking about amateur rather than professional photography, even if we're simply talking about interpretation, it's because otherwise photographs would all be structurally the same; a daytime photograph at ISO 100 is the same as a nighttime photograph at ISO 6400. This didn't happen in analog photography, de gustibus.

This is photography today, software-assisted, with cutting-edge technology. But we shouldn't forget that for the camera sensor, however modern, noise is physiological—can we say physiological? Let's face it. It is just as physiological as it is for analog photography.

As the 100% crop in example #3 shows
Perhaps part of the problem is monitor viewing. I'd be inclined to believe that if digital photography didn't go through the monitor, but directly to print, photographers wouldn't make such a big deal out of it.
At least up to the nominal print magnification. For higher magnifications, as you can see from this example, the situation would be different.
This doesn't mean that digital noise isn't a problem. In some respects, or in some cases, film grain was too.

It simply means that perhaps it's better to evaluate digital noise more carefully, and with less prejudice, transforming it into an interpretative opportunity, where the defect, at times, is an added value, which can give the image a more romantic and analog look.

For those amateur photographers who like this concept, I'd recommend starting by trying to eliminate only chrominance noise, thus achieving a nice grainy, analog look. Try it and see how beautiful it is.

Because, let's face it, unfortunately, digital noise is different from film grain; it's certainly more unpleasant, but it remains intrinsic. An interesting aspect is that you can work on luminance or chrominance noise without distinction.

Precisely because it's unpleasant, in many cases, noise reduction is necessary. And we who use Color Perfect must find the best way, the method, to eliminate digital noise from our images.

This discussion isn't meant to be a noise-positive versus noise-negative debate. We're all aware, for example, that a nighttime cityscape can look unsightly at ISO 6400. But it also depends on the intended use. And in any case, many other photographs may not have the same unsightly problem.

So, if we want, or need, to eliminate digital noise, I'd say there are two methods:

The first is to plane it after finishing the processing with Color Perfect, so once the work is finished. And there are many software programs for doing this.

The second is to plane it before using MakeTiff, that is, first run it through a software that cleans it up and returns a linear DNG file to then pass through MakeTiff.
Here, however, it seems to me that not all software is suitable, because the conditions are to have a linear DNG file returned.

I don't know which method is best.

I'd assume the second one.

Since I don't have software that can output linear DNG, I was only able to experiment with the first method—cleaning it up after the job was completed—and I have to say it can be done; I'm happy with the result.
This is coming from someone who also likes digital noise, as long as it's not too excessive! So the discussion is open.

The 100% crop in example #4 shows the file first processed with Color Perfect, and then planed of chroma noise only.
The 100% crop in example #5 shows the file first processed with Color perfect, and then planed for luminance and chrominance.
Examples 6 and 7 show the same processing as the respective examples 4 and 5, but with a crop to the nominal print size. That's the maximum enlargement, of maximum quality, that you would get at 300 ppi with the camera in question.
It should be noted that in examples 4, 5, 6, and 7, digital noise reduction was performed using Dfine2 from the first version of Nik Collection, which now uses outdated technology. Therefore, these are not intended to be optimal examples, but rather conceptual examples.

In any case, leaving aside the fact that my noise reduction is poor, and that current software can achieve miraculous results, my favorite images are 2 and 6. While 2 probably still shows some chrominance noise during printing, but perhaps you should look for it with a magnifying glass; at the very least, it should be much less noticeable, and therefore, perhaps still pleasing. And 6, in my opinion, is spectacular with this "dirty" and grainy effect that reminds us of analog photography.

So the message is: if we use high sensitivity in our digital cameras, then let's show it off, at least when it's worth it. After all, that's what Color Perfect users do when scanning or reproducing from negatives. For all other cases, personally, I'm looking for the best and most natural method for noise reduction.

As I said, I assume the best solution would be to reduce the noise before processing with Color Perfect, but I don't have any examples to show, and I don't know if this is actually the case.
If anyone has done any tests on this and knows the best solution and method, it would be helpful and interesting to know.

All examples have been sharpened using a sharpening mask (not Nik Collection Sharpener Pro 3) to show a roughly complete result.

Beautiful frame. Can we proceed looking at both venues with this very image, or if "we" cannot, can I
Comparing the pre PerfectRAW options would be interesting.
I'll find or choose another example image otherwise, no worries

Of course!
That's why I'm here.
Here's the Raf.

First, let’s look at what we have in Photoshop CC. I don’t have anything newer than the 2023 version on this Mac, but it will do for a first look.
I believe in later versions this has evolved into something else. We’ll take a look at that subsequently.

• We open your .raf in Adobe Camera RAW (not to process—why would we).
• We just want to access one of the hidden built-in features: “Enhanced.”
• We right-click the image in thumbnail view to find that.
• In the menu that comes up we select Denoise and set an Amount.
• We press the Enhance button and the attached DNG is computed.

You can make TIFFs out of this and Roberto’s *.raf using MakeTiff. Do whatever you like to the denoised version, then open the noisy one and on the Start panel press the button controlling the carrying over of values until it reads Exact > As Optioned. The entry before the arrow is where we are (“Exact”), and the one after the arrow is where another button click takes us—and that cycles around.

Copy them above each other in Photoshop as layers and compare; you can also blend them easily this way. The denoised image is too clean, but it is impressive nonetheless. Personally, I would much rather add something that resembles the beautiful grain of a black-and-white film than retain the high-ISO noise pattern we started with—but that is a matter of choice.

That's what I say too, why should we process it with PS?

Amazing, it's a great job compared to what I can do with obsolete software.

Let's see if I understand correctly:
Once I have the noise-free DNG, I run it through MakeTiff and obtain the linear TIFF.
I process this TIFF.
Then I run the linear TIFF with digital noise through Color Perfect, and by pressing the Start button, I set it to: Exactly > as Optioned.
This essentially applies the changes made to the previous TIFF, the one with noise reduction.

After that, I stack the two processed files on top of each other in PS, and the opacity serves to increase or decrease the intensity of the noise reduction effect.

A great solution!

Evidently, choosing analog grain is better than choosing digital noise. For me, the important thing is to dirty up the image, and grain is undoubtedly the most beautiful.

The best grain, however, is not the one you can achieve digitally through software, although I must admit there are some very interesting solutions.
Rather, the best grain is the one we can get from analog film, through a reproduction or scan of the negative and then apply it to our digital file.

We can obtain grain from different films: black and white grain, and colored grain from color film.

The problem is creating them well, with quality tools.
The point is that I personally don't know how to create them.

I have grain from many films: Kodak, Fujifilm, Agfa, Ilford, Rollei, and others. The ones I use most are grain from Kodak Tri-x, Tmax 3200, Kodak Vision 2, Kodak Portra, and Orwo N74. The nice thing is that you're adding the grain of color film to a color file.

These files are the right size for my Fujifilm X-Pro2, so they're a good size to allow for not over-enlarging for larger files, and not over-reducing for smaller files.
It's basically an 18% gray file with the grain on top, applied to the file with the Overlay blending mode for greater intensity, and with the Soft Light blending mode for less intensity. Then you can further adjust it with the opacity setting.

The point is, they're not perfectly made; they're beautiful, but you can do better with the right technique and equipment.

For example, many of my files are slightly darker on the top, others are slightly darker on one side, so you have to try different flipping solutions to prevent an already dark area from darkening even further. Essentially, they don't have a uniform light.
I could understand a vignette, if you wanted; it would be appropriate, but this way it's a bit boring.

So you should take them with even lighting and a lens with very little vignetting.

That said, I use mine with great satisfaction; I use them on all my photos, and they're wonderful.
I also use them over digital noise, because they're not like the digital grain created with software. It's definitely less invasive; it's not coarse or coarse, it's not as visible as digital grain, it's more delicate, and above all, it's natural. It's practically as real as what you see on analog film; it's almost the same.

We need someone here who knows how to do this kind of work and has the desire to do it.

I personally would commit to purchasing the films.
If anyone is interested in this grain, we could collect a dozen or so rolls.

Personally, I'd even be willing to pay for the trouble of this work, so much is my interest in this type of solution. But it has to be a job well done!

That said, I'm willing to share my film grain files with anyone interested.
In any case, these files are available online for free.

Yes, exactly. The route detailed creates a DNG that can be converted to a linear TIFF like any other using MakeTIFF. That is also true for the Super Resolution or Enhanced Detail files that are created from images at base ISO.

In this case here, I would do my editing work on the noise-free image in ColorPerfect and then carry over its settings via Exact to the noisy image. The reason is simply that with the absence of noise, nothing can throw off BP tails, etc. I will follow up on the analog grain insertion later.

It may be worth pointing out that there is more at play here than the simple act of “dirtying up” a clean frame. Film grain has an effect that goes beyond style or nostalgia. It acts a bit like the dither that is deliberately added in other fields of signal processing.

If you think about a smooth gradient that is being held back by the finite steps of digital encoding, or about shadow tones where too many values are crammed into too few levels, you can see how things may break down into banding or flat patches. When a very fine, irregular grain is present, the boundaries between those steps are no longer rigid. Different parts of the image fluctuate ever so slightly across those thresholds. Our vision is then able to piece together distinctions that would otherwise collapse into sameness.

So paradoxically, adding that delicate, irregular texture increases the number of tones we can actually tell apart. What would have been blocked up becomes nuanced again. That is why high-ISO film, when well printed, still looks rich in its shadows despite its grain, and why the right kind of film-derived grain can make a denoised digital file feel more lifelike than the mathematically perfect but sterile version.

This does not need to be pushed as far as theory. You can demonstrate it to yourself by preparing two test files: one a smooth dark gradient, the other the same but with a whisper of grain added. Print both at a decent size and the one with grain will usually show you more subtlety rather than less.

That, I think, is the deeper reason why this path feels satisfying. It is not merely about evoking the past, but about unlocking something that is there in the image but cannot come across without that tiny bit of texture to guide our eyes.

Good for the path to follow with the file cleaned of digital noise.

I've always thought that any interference on the file, such as digital noise, altered the file's computational evaluation, but I believe this isn't unique to CP.

I'm aware of your explanation regarding analog grain.

My desire, in fact, stems from giving that analog touch to digital photography—in other words, making it more textured.

And I immediately realized the important value it brings.
I couldn't have explained it like you, but it's immediately evident; it benefits the shadows, but it benefits all areas of the image.
Perhaps this effect gradually disappears as you reach the highlights, although you still benefit from it.

One more thing:
At this point, should the cleaned DNG to be passed into MakeTiff be demosaiced with LibRAW, and not with AdobeDNG?
And this should also apply to the original, uncleaned RAW.

While we're on the subject, the same applies to cameras that generate DNG, like the Leica and Ricoh. Do we go directly through the LibRAW engine?

robyferrero wrote: Mon Sep 15, 2025 12:32 pm At this point, should the cleaned DNG to be passed into MakeTiff be demosaiced with LibRAW, and not with AdobeDNG? And this should also apply to the original, uncleaned RAW. While we're on the subject, the same applies to cameras that generate DNG, like the Leica and Ricoh. Do we go directly through the LibRAW engine?

Historically, the results obtained with Adobe’s DNG Converter have always been superior. With dcraw it was, in fact, impossible to extract data from color negatives that would invert properly. LibRaw, however, with its addition of the DHT interpolation method, finally brings a mode that is equally well suited. I even have examples shot under a light source far removed from daylight where Adobe’s DNG Converter fails, but those are edge cases that can be discussed separately.

There is one very simple reason why the DNG Converter will remain the number-one choice going forward. Virtually all mirrorless camera systems rely on lens designs that require geometric de-distortion as an integral part of the imaging chain. The correction instructions are embedded in the raw data. Adobe’s DNG Converter reads and applies them; LibRaw does not. And because these corrections are very difficult to reproduce later on — and in practice there is almost never a need to — having them applied right at the start is a decisive advantage. The fact is that nobody produces lens profiles or post-processing tools intended to recreate this step afterwards; all existing profiles assume the in-camera or DNG Converter corrections have already been performed.

There you have it!
I'm using the LibRAW engine because it seems to deliver more detail and a greater perception of sharpness.

So the same goes for native DNG files like Leica and Ricoh; it's better to process them through AdobeDNG.

Examples of noise reduction performed on the Raf file using CameraRAW, creating a Linear DNG file to upload to MakeTiff and processed with PerfectRAW.
In example #2, you can see the difference compared to the original file without noise reduction. The difference is notable, even in terms of color balance, where, with CameraRAW's Linear DNG, it was easier to achieve a better color balance. Managing the entire process also seems easier.
The result also appears to be a better balanced image across the entire curve, from lowlights to highlights.
Example #3 is a 100% crop of the nominal print size.
Example #4 is a 100% crop.

I think you should elaborate on how you put the noise back in for artistic reasons, how much and so forth.
Can you also post a *.CPMetaR file of the edit you like on the denoised file? I'd like to follow up with some grain examples.

I haven't saved the Meta file, and unfortunately I can't save it because I've already processed another file.
But it's not a big problem, as I can redo the process the same way and save the Meta file.

Regarding the noise visible in CameraRAW's Linear DNG, I was a bit perplexed, as I didn't add any noise. What we see is what's in the Linear DNG file, which is still much better than the original file without noise reduction. But there is still noise.

This huge difference in color perhaps proves what you said, namely that Color Perfect's calculation is fooled by digital noise.

Consequently, I think that every tiny trace of noise in a file should be reduced beforehand.

But in this case, we're talking about a file at ISO 6400. If instead it were a 400 ISO file, where noise is present, even if extremely limited, for Color Perfect to work well, would pre-noise reduction be necessary?

You must have mixed something up, the file I have open is soft as a baby's bottom... So some granularity must come back

But in fact, I remember your words about smoothing the file; you said it was very smooth.
I demosaiced the Linear DNG with AdobeDNGconverter, and then with CP I adjusted the exposure, color, gamma, graduated white, and highlights. That's it, that's what came out.

This is the Linear DNG opened with my CameraRAW.

robyferrero wrote: Tue Sep 16, 2025 3:54 pm with my CameraRAW.

Why that, does it come out smoothly from MakeTiff with DNG Converter, is libRAW different?

I mean, I wanted to view the file in CameraRAW to see if I hadn't made any mistakes when converting to MakeTiff and processing with CP.

But as you can see, the Linear DNG file I have is dirty, regardless of the processing in MakeTiff and CP.

However, when I demosaiced the Linear DNG with LibRaw, it didn't demosaice. To do this, I have to use Adobe DNGconverter.

We'll figure it out for the time being I'll post the Tiff file.

This is the file corrected with CameraRAW noise reduction, converted to Linear DNG, then demosaiced with MakeTiff and processed with PerfectRAW.

You can still see another color.
This is the original file without noise reduction.
I'm posting it to show the difference in color, even though I performed the process by loading the "Exact" data.
I hope I followed the "Exact" process correctly,
also because, in reality, in addition to having a slightly different color, there are other values ​​that need adjusting.
I don't know if this is normal, but I think so.
This example shows the crop at 100% of the nominal print size.
This is the same example with the addition of analog grain from Kodak Portra film.
This example shows the 100% crop.
This is the same example with the addition of analog grain from Kodak Portra film.
It should be noted that the analog grain was added after sharpening the image. Personally, I prefer the grain to be less sharp.

Despite this, it's likely that the image sharpness, in this specific case of noise reduction, could be improved.

I didn't merge the two files, the clean and dirty ones, in Photoshop. This is also because I wanted to understand more about this color difference. I'm trying to figure out if I did something wrong, or if it's normal.

I'm attaching a CPMetaR file.

Again, I hope I've done everything right when creating the CPMeta file.

The meta file is perfect — little can go wrong. It’s only about having the box checked. If it is checked and there already is a file, a save dialog will pop up. Canceling out of that only cancels the file save operation; image processing continues.

Now, I see the color difference you mean, and I don’t like it. It has always been the case that the removal of color noise also takes a good portion of the actual color with it.
We have more modern versions of the process to check — 2023 was a while ago.

Anyhow, we are well equipped to deal with that.
With your meta file, I applied the exact same settings to both the regular and the optimized file, and on top of that I interpolated with LibRAW and DNG Converter.
Now, with the proceedings detailed here, which are simple and fast once learned, we can preserve the color of the non-“enhanced” version while applying the exact brightness values of the “enhanced” image. That effectively removes all luminance noise from either version and leaves only the chrominance noise, with exactly the same color.

There is a difference between the two versions, but for X-Trans sensors LibRAW looks pretty bad. You will find color artifacts, and those are what make the difference.
So the last step would be to re-apply some finer granularity. I did something quick by emulating analog Tri-X Pan grain, as if this were shot on 120 film. I could have done much less, but I wanted something visible.

So, we’ll test newer pre-processing denoising options, but for now — and maybe for good — we can praise the Alpha feature of ColorPerfect.

Yes, I know, LibRAW offers more detail, but also more artifacts on X-Trans.
All the layer examples are clear and perfectly illustrate what you explained. I'll go back to using AdobeDNGconverter.

The Alpha function for performing this type of operation is very interesting; I'll try it.

But I'm also wondering if it's possible to start with an intermediate noise reduction level directly in CameraRAW.

The Tri-X Pan grain is beautiful, and, as you said, it really does a useful job.
I also keep mine less visible, and in this case, it's at 100% to make it more visible.

Next let's go commercial. DXO PureRAW 5 with its new X-Trans specific stuff.