I try to improve the original but I am not satisfied; any suggestion is welcome thk
original


my tests

I tried blue channel extraction set to overlay, but the noise was just too much to bother sharing the result here Marcello (and I'm too lazy to do cleanup work; lol). You can, if you are willing to address the noise with some tlc work, try to check out various channels to see which one makes sense to enhance the wisteria.

Open image
Colors>Components>Decompose (HSL)
on hue layer: Free-hand Select blossoms, Invert selection, Bucket-Fill dark grey background (use Color picker eye-dropper to get a good grey matching the area immediately around the blossoms)
this gives a mask like this:

Select all, copy and paste as new layer over original image
Duplicate original, set new layer (ie. the one just pasted) to multiply mode, merge down and set resultant layer to addition mode
Duplicate this layer and adjust opacity to taste:

Here's my go. Decompose to LAB, used same curves on A & B, Unsharp Mask on L.

Thank you friends..you are fantastic!

Nice result jontait2.

Thanks Lyle
Actually I think the best result would be somewhere between my effort and Odin's.
Perhaps overlay the two and blend!

50:50 blend of my result with Odin's:

jontait2, The combined result looks pretty good.

my try:
- use of perspective tool (moderate)
- use of NL filter (moderate values)
- use of Richardson-Lucy sharpen (GMIC)
- use of Extract bright colors (GMIC)
- overlay, merge

simpler procedure:
-duplicate overlay
-duplicate, invert luma, overlay

use of my personal GMIC filter on tone mapping (quite complicate to describe)

Lylejk, jontait2, odinbc, dinasset and sallyanne a big thank
I will try to learn from your suggestions

Just to update

I succeded on jontait2 procedure(decompose HLS , made a mask with the Hue layer etc)

On the OdinBC procedure I learned the decompose LAB ( also reading the article of Helen Bradley in digital-photographic-school.com) but my big problem was to find the right curve to apply to layers A and B!!

Dinasset first procedure stopped me because I didnt find in my GMIC "Extract Bright Color"??!!

Instead I found the second procedure to be very simple but great result

I was very close to the Sallyanne result that I find very interesting

Thank you again for your replies and for the tests that you made; I learned a lot!

Your original image curves after decompose to LAB.
B has the widest histogram at x200. I then use 55 + or - on both A & B as the magic number, (a safe choice).



Note the large difference between the histogram edges with about 95 being the next edge. In this case using up to 70 as the magic number on both A & B should give acceptable results. Upping the number to 70 would not be a good choice if the widest edge of B histogram had been (unlike this one), a steep initial slope.

Sorry, Marcello, you're perfectly right, with THAT NAME it doen't exist under G'MIC.
That's because I copied it to my favs and changed the name (the original name didn't say much to me), the original link under G'MIC is the following:
TESTING
- GENTLEMANBEGGAR
-- MAXFIELD PARRISH
Now you may try also that one...

The problem is underexposed object with bright background, so you will need to separately adjust the curve. jontait2 suggested a way to create a mask, but for this image I would choose an easier way by using select by color. Then gauss blur the mask with radius of 2. Then I push the top input level down in the Levels tool to brighten up the blossoms. This way you can see how well your changes blend into the background without any guessing work. Finally add the saturation, sharpness, local contrast enhancement to the whole image.

Thank you Marcello for your nice photo and sharing it with us.
Great improvement made for all of you.
Here it's mine.

second update

Dinasset - I found the filter! my result is almost similar to yours Thank you

OdinBC - with your curve data I was able to have a result around yours

Okyl168 - for me was difficult select by color ( did you set select by Hue? with threshhold 45 ?); other question : did you use MathMap>Colors>local contrast enhancement?
here is my result