Understanding the under pinnings of Raw Processing.

Being a huge nerd who loves low level underpinnings of software and technology I have been trying to Google Foo some technical information and white papers on the nuts and bolts of processing raw files. Sadly I can’t seem to find anything. Maybe me search criteria is totally wrong. Can anyone help point me in the right direction?

I have seen basically no resources online either. Everything I’ve learned to make Filmulator has either been by reading source code of other editors or by experimentation.

1 Like

It’s a very technical stuff and I find it hard to get good in-depth information.

Some good resources I found:

You can check the video in the first post of this thread:

Here is something about noise and where it comes from:
Shot noise

Electronic noise

This is the most sophisticated abstract about noise I found so far, very scientific

1 Like

That’s from Emil, who made Amaze demosaic, Raw CA correction and some more stuff.
Though I optimized his stuff for speed, I have to admit, I still don’t understand it in whole…

1 Like

Not quite sure what you’re after specifically, but I like these basic hands-on guides here and here (this blog has great follow-up posts as well).

to get you started on some technical aspects i recommend this lecture:



Me neither, especially the math…
But I was able to extract some valuable contexts out of that work between topics noise, exposure time, ETTR and ISO. Only found articles elsewhere about these topics in separate without getting the big picture.

For example, you use a higher ISO setting of 3200 to expose-to-the-right in a darker situation instead of going for ISO 100 and pushing the exposure in the RAW developer later.
Since you don’t have to push the exposure very far with ISO 3200 it could be that the stronger brightened ISO 100 image is the one with more noise.

This one is a good start:

1 Like

That depends on the sensor technology. Now, some sensors have ISO-invariant noise, and since high ISO sensitivity is achieved while reducing the dynamic range, you would better shoot at the lowest ISO available, and figure out noise vs. shadows recovery later. You need to check for your particular sensor.

1 Like

That’s a big topic. There are two main phases: (1) reading the raw file, and making a denoised, color-balanced and demosaiced image, then (2) prettifying that by adjusting tones, sharpening and whatever is wanted. Phase (2) may be automated or manual or both.

Technical information is available on the components of each phase. Generally, search for “image processing” plus other terms.

Thanks for all the information. Hopefully some of this will help me understand better what is going on under the hood of the software. As I find this stuff fascinating I doubt I would ever be able to do my own little fun personal project but my cog wheels are turning and I just have to satisfy them with information even if it ends up way over my head.

An excellent reference. And one can go waaaay deep down the rabbit hole by following his work since then. He was one of the co-authors of Google’s HDR+ paper in 2016 ( https://hdrplusdata.org/hdrplus.pdf ) - parts of that pipeline have been replaced by https://arxiv.org/abs/1905.03277 - also work LeVoy was involved in.

(I’ve used Tim Brooks’ implementation of the align-and-merge portion of the 2016 pipeline with bursts from an A7III with highly promising results.)