Histogram equalization better in linear colorspace?

[vicmortelmans]

Hi,

I'm not an expert, but I'm also busy developing my own negative inversion algorithm (in the beginners phase). One of the key elements, as I understand, is to do the leveling operations in linear colorspace.

I've eagerly downloaded your neat application and tested it on a number of raw digitized negatives, and then looked at your code and modified it to eliminate all non-linearity in the raw conversion and (at least on my scans) this seems to improve the colors a lot!

I also learnt (and tested) that using an (am)oled display as backlight for digitizing gives better color separation than a full-spectrum light, because of the specific red, green and blue frequencies it's emitting. If you confirm this, it may be a good hint to put in the README!

Here are the changes that I made to make the raw conversion linear:

diff --git a/source/RawProcessing.py b/source/RawProcessing.py
index 4418742..c514757 100644
--- a/source/RawProcessing.py
+++ b/source/RawProcessing.py
@@ -20,11 +20,11 @@ class RawProcessing:
     frame = 0 # adds white frame to final photo
     jpg_quality = 90 # from 0-100
     tiff_compression = 8 # defines the tiff compression algorithm
-    dm_alg = 2 # demosaicing algorithm
-    colour_space = 7 # output colour space after demosaicing
+    dm_alg = 0 # demosaicing algorithm
+    colour_space = 0 # output colour space after demosaicing
     exp_shift = 3
     fbdd_nr = 0 # noise reduction on demosaic
-    raw_gamma = (2.222,4.5) # (power, slope) for RAW image
+    raw_gamma = (1.0,1.0) # (power, slope) for RAW image
     use_camera_wb = True
     wb_mult = [1, 1, 1, 1]
     black_point_percentile = 0 # sets the black point

This is what I was recently reading on this topic:
A Better Light Source For Scanning Color Negative Film and Tri-Color Scanning, Color Negative Film & Color Spaces

Thanks for your contribution to analog photography!
Best regards,
Vic

[kaimonmok]

Thank you for the suggestion. I'm always open to having a second opinion on the inversion algorithm, so if you ever finish developing your inversion algorithm, and would like to share your findings, please do! And I can certainly add using an OLED to the README.

Here are some of my comments on the issue of colour spaces:

1. dm_alg definitely has no impact on the overall colour, except at a pixel level. Unless there's a specific reason to go with a certain demosaicing algorithm, I don't think there is much need to change this.
2. Yes, setting the gamma to (1, 1) will indeed output in linear. Setting colour_space = 0, which is the 'raw' colour space, I'm less sure about. In my testing, it produces very flat colours, which may be more desirable for editing, but as for producing a pleasing output, I think we would need to convert it back to normal colours.
   I'm not sure what the exact solution to this is; hopefully somebody with more knowledge in colour spaces can chime in. I think more research into the rawpy/libraw docs is necessary to establish the ideal settings for film scanning.
3. Edit: I would also add that exp_shift = 8 (+3 stops) has a very big impact on colours. For some reason, boosting the exposure makes the colours look better, but I don't understand the objective reason why. There are many other settings that I haven't included from the rawpy docs that may or may not impact the colours also, if you would like to look into it further.

Let me know your thoughts!

[kaimonmok]

So I've been playing around with this problem a little bit. For me, using a linear colour space has mixed results. In general, I get a much lower brightness on the raw scan, and colours look a lot flatter in the final image. I think there needs to be some conversion to bring it back into sRGB colour space, so I've been trying to reverse engineer how the gamma parameter actually works.

I've gotten as far as finding the method that deals with gamma in the LibRaw source code, but documentation is extremely sparse. All I know is that the two values represent power and slope parameters, but I can't find the exact formula spelled out anywhere.

For now, if those settings work for you, then stick with it. But I don't think I'll change it in the main branch until we figure out definitively what's going on.