Commit Graph

15 Commits

Author SHA1 Message Date
rom9
22eee9787e Film negative stable multipliers (#5485)
* Added new feature to calculate channel multipliers from crop area. This also saves the crop area channel medians to the processing profiles, in order to get a more consistent color balance when batch-applying the same profile to multiple pictures from the same film roll.

* Fixed wrong initialization of array, and missing check for the result of `getFilmNegativeMedians()`.
Moved `ImProcCoordinator::translateCoord()` from private member to anonymous namespace.
Fixed some whitespace and formatting issues.

* Fixed some formatting issues

* Passed `ipf` parameter as const in `translateCoord`.
Narrowed `using namespace` to single class `Coord2D`.

* Added `scaleGain` entry to thumbnail metadata file, to make `scale_mul` multipliers available in thumbnail processing phase. This simplifies multiplier calculations, so that "faking" thumbnail multipliers in the main image processing is not necessary anymore. This way, output values are immune to slight variations of multipliers between successive shots taken with in-camera AWB turned on.
Shifted multipliers so that the output channel medians are balanced when "Camera WB" is selected. This way, just computing multipliers from crop and setting "Camera WB" (which is the default) gives a pretty well balanced image as a starting point.

* New channel scaling method, based on a film base color sample instead of crop area channel medians. Channels are scaled so that the converted film base values are equal to 1/512th of the output range (65k). This giver better black levels in the output, and more consistency when batch-processing multiple negatives.
The output is now compensated for a known fixed WB value, so that the film base will appear grey when WB is set to 3500K, Green=1.
Added PPVERSION 347 to preserve backwards compatibility: when a processing profile saved by RT 5.7 is loaded (PPVERSION=346), the new fields are initialized to the special value -1, which will instruct the main processing to follow the old channel scaling behaviour. The old channel scaling multipliers will then be converted to the new film base values so that the resulting image is the same, and the fields will be overwritten as soon as the PP is saved again. This will transparently upgrade the processing profile.
When the new behaviour is used, but the film base values are still unset, they are estimated based on channel medians, excluding a 20% border around the image. This should give a better result out-of-the-box for pictures containing a large film holder.

* Code cleanup from review

* Run astyle on film neg source files

* Fixed automated build failure caused by incompatible libraries on my dev PC.

* Simplified `Thumbnail::processFilmNegative` method signature. There is no need to pass in `rmi`,`gmi`,`bmi` multipliers from the caller, i can do the same with my own internal multipliers.

* Added `FilmNegListener` class to pass estimeted film base values to the GUI after first processing. Removed old `filmBaseValues` instance variable from RawImageSource.

* Code cleanup

* Forgot to set baseValues flag in `PartialPasteDlg::applyPaste`
Fixed `filmBaseValuesLabel` not updating when reading zero baseValues. Normally not needed (the label is updated later by the listener), but when the user is browsing through pictures the listener won't fire, so the label must be updated to show values are unset.

* Overwritten channel scaling multipliers by calling `get_colorsCoeff` with `forceAutoWB=false`.
Initially, in `RawImageSource::load`, channels are auto-balanced by averaging the whole picture when computing multipliers.
This can give different multipliers for multiple shots of the same camera, which will lead to inconsistent conversions when batch-processing multiple negatives.
This commit re-sets `scale_mul`, `ref_pre_mul`, etc., in order to "undo" the auto-WB and use the normal camera multipliers.

* Found an easier way to get stable overall multipliers, removed the (horrible) on-the-fly mutation of scaling instance variables.
2020-04-13 17:20:56 +02:00
Ingo Weyrich
587652283d further reduction of include dependencies, also finalised some classes 2019-11-09 16:51:54 +01:00
Morgan Hardwood
60b2196bce Updated gnu.org links to use HTTPS 2019-09-10 12:34:57 +02:00
Ingo Weyrich
46cbb652d5 cppcheck: further fixes 2019-08-01 14:02:38 +02:00
Ingo Weyrich
0493e67474 nuke empty adjusterAutoToggled functions 2019-07-20 11:44:38 +02:00
rom9
63a6e1b79c Changed GUI behaviour so that the first slider sets the reference exponent (green channel), and the other two sliders set the ratio of the red/blue exponent to the reference one. This should be more intuitive, and less confusing when clicking on the reset-to-default slider button. Now the first slider sets the general contrast, while the other two affect the color.
Lowered exponent slider max value from 6 to 4 (it was too large to be useful). New sliders are log-scale.
2019-06-26 02:00:34 +02:00
rom9
8c6458daff Removed "lock channels" CheckButton, it wasn't very useful. Keeping the "(lead)" note on the green channel label, to give a hint that this slider controls the other two as well. 2019-06-21 20:51:50 +02:00
rom9
7c0275ca1a Updated copyright notice in new source files 2019-06-18 13:52:58 +02:00
rom9
8de581ac39 Removed old debugging stuff from rtgui/filmnegative.cc 2019-06-17 21:46:50 +02:00
Flössie
b687ca2643 Merge branch 'rom9-filmnegative' into filmnegative-cleanup 2019-06-17 08:57:14 +02:00
Flössie
54cc02eea9 Filmenegative core cleanup 2019-06-17 08:03:46 +02:00
rom9
d01f78864c Locked red and blue exponents by default in the GUI, added CheckButton to unlock them. 2019-06-14 22:14:24 +02:00
Flössie
80f2b6a002 Replace raw arrays with std::array<> 2019-06-14 08:58:04 +02:00
rom9
b95bdb1aea Linked red and blue exponent adjuster to the green adjuster, in order to maintain the ratio between exponents if the user moves the green adjuster (master).
Switched back to vector sort for median calculations: the results of the histogram search function diverge more and more from the simple median calculation as the exponents increase.
At 2.0 the test picture is already impossible to WB as the multipliers are too far off (2.78226e+08 histo vs 9.7927e+11 sort), and the normal WB sliders can't compensate for those huge factors.
2019-06-10 22:05:54 +02:00
rom9
9df8008949 Film negative processing: first usable version. Only supports bayer raw files, thumbnails don't work
Added performance improvements suggested by heckflosse. Lowered median sampling step from 7 to 5 since calculation is now much faster.

Added support for Fuji X-Trans raw files.

Applied SSE2 patch provided by @heckflosse, improves performance in main processing loop.

Moved film negative processing stuff in its own compilation unit.

Code cleanup: removed redundant omp directives.

Added check for dead pixels, going above threshold after inversion. ST_BAYER only for now.

Reverted leftover hack in cropwindow.cc
2019-06-08 01:40:29 +02:00