Let's improve grain


I like the proposed darktable noise. How does it compare to the current implementation at other “ISO” levels?


(Andrea Volpato) #42

To spot some differences one could look at the lightness histograms of the grain for 10%, 50% and 90% gray.

The darktable grain at 100% intensity has the following histograms.
The 10% and 90% histograms look a lot narrower. I wonder if it would be nice to tweak them so that they match your grain @Carmelo_DrRaw, which is more sophisticated.

It performs essentially in the same way since it is only a tweak of the weights and frequencies of the noise octaves.
Here are the power spectra trends.


For direct comparison here is the old implementation.


(Andrea Volpato) #43

@Carmelo_DrRaw, can I ask you a favor?
Could you apply the film grain of your implementation to this image:

I’m interested to see the full lightness profile of the grain.


(Carmelo Dr Raw) #44

Here is my result:


I can provide you results with other grain densities if needed…

1 Like

(izara) #45

First of all, I think your work is very impressive, thank you for sharing it! Can I ask you a question? When you say:

“The light filtering trough the photographic film will be affected uniformly and we need to add the noise to the logarithm of the exposure.”

Why do you add the noise to the logarithm of exposure and not to the density of the film? I’m just wondering, maybe I don’t completely understand how the process works, because if the grain is in the film it should be added to the film, isn’t it?

And also, when you say uniformly you mean the noise follows a normal gaussian distribution?

Thank you in advanced!


(Alessandro Amato Del Monte (Aadm)) #46

Experimenting a bit with grain and thinking “well that’s an area where Lightroom still has an edge” (but honestly, I haven’t double checked and my memories may be a bit hazy; I should’ve made a comparison but tonight I’m lazy).

Then I’ve found this thread and wow this truly deserves an article of some sort and to be put on the front page of the NYT or something.

Anyway, I’ve understood perhaps 1% of all the math above, but what I did understand is the meaning of the midtone bias slider in the grain module. I think the effect is really good and realistic:


right: grain 1600ISO/strength 50% / midtone bias 100%, left with midtone bias 0%

Apart from the blog post or article mentioned above, I also think that the online manual should be updated because as of now, it does not mention what is the midtone bias slider.


Denoise, DT versus RT