Example: Left neutral profile, middle neutral + raw ca correction + avoid colour shift, right same as middle, but with patched code

diff --git a/rtengine/gauss.cc b/rtengine/gauss.cc
index ab080a3c4..b7de67851 100644
--- a/rtengine/gauss.cc
+++ b/rtengine/gauss.cc
@@ -1143,7 +1143,7 @@ template<class T> void gaussianBlurImpl(T** src, T** dst, const int W, const int
{
static constexpr auto GAUSS_SKIP = 0.25;
static constexpr auto GAUSS_3X3_LIMIT = 0.6;
- static constexpr auto GAUSS_DOUBLE = 70.0;
+ static constexpr auto GAUSS_DOUBLE = 25.0;
if(buffer) {
// special variant for very large sigma, currently only used by retinex algorithm
Unfortunately the patch does not solve all cases. For example this one still shows the issue:
Left neutral profile, middle neutral + raw ca correction + avoid colour shift, right same as middle, but with patched code

Using the better double precision gaussian blur increases processing time by ~10% in case avoid colour shift is enabled. If there are no objections, I will push the patch asap
@heckflosse
Where does the pink cast in the clouds is comming from?
I can't really see a different between the middle and right image.
@TooWaBoo The pink cast is a result from 'avoid colourshift'. There is quite heavy ca (or fringe) whichs correction leads to a shift of the average colour in that regions, which then leads to a wrong correction by 'avoid colour shift'.
I showed the example to show, that the issue is not solved for all cases. You can't see a difference between middle and right in last screenshot because there is no difference.
@heckflosse
Ahh, I thought I'm blind. 馃槄
No objections to merge.
Most helpful comment
Using the better double precision gaussian blur increases processing time by ~10% in case avoid colour shift is enabled. If there are no objections, I will push the patch asap