Noise reduction fails when image contains a large uniform (i.e. overexposed) area

Beep6581 / RawTherapee

A powerful cross-platform raw photo processing program

https://rawtherapee.com

GNU General Public License v3.0

2.93k stars 326 forks source link

Noise reduction fails when image contains a large uniform (i.e. overexposed) area #3469

Open heckflosse opened 8 years ago

heckflosse commented 8 years ago

From here: https://discuss.pixls.us/t/noise-reduction-for-highlight-jpg/2301/11

The image contains a very large overexposed area which leads to noise reduction failing in preview and in all modes except Auto multi-zones.

I pushed a fix to a new branch. I also checked before/after fix with a lot of files which don't contain a large uniform area. For these files I detected no differences visible by eyeball. Only differences using amplified absolute Difference.

Tests are welcome,

Ingo

Beep6581 commented 8 years ago

heckflosse commented 8 years ago

@Beep6581

Related: https://discuss.pixls.us/t/image-in-viewer-different-from-rawtherapee/2521

No, it's a different issue. This one is related to Denoise Luminance-Detail (Detail recovery)

Hombre57 commented 7 years ago

I'll let you compare the result of PR #3470 against the dev branch here : https://filebin.net/1c97emtkw8yhtb7s

Overall, as you can see at the top left of this image, small saturated colors are slightly more smeared, but I don't know if it should prevent @heckflosse from committing the patch.

heckflosse commented 7 years ago

@Hombre57 I see what you mean. Maybe Jacques can chime in and test when he's back from vacancies @Desmis. The patch is so old that some additional weeks won't matter at all :)

heckflosse commented 7 years ago

@Hombre57 I compared before/after of your file at 200%. To me it seems that with my changes the nr is a bit stronger without loosing relevant details. But that may be different for other files. We have to test further. Maybe @Beep6581 can also chime in.

Desmis commented 7 years ago

My grandsons have left since Sunday, and I am going to the Caribbean next Saturday...for 2 weeks :)

I looked at the patch proposed by Ingo and it seems to me that it provides a good solution to the problem. Maybe there is another solution, but at the moment I do not see which. Thanks to Ingo :)

heckflosse commented 7 years ago

@Desmis

and I am going to the Caribbean next Saturday...for 2 weeks :)

Lucky you!

Thanks for looking at my patch :+1:

Ingo

Desmis commented 7 years ago

Ingo

Thank you :)

Hombre57 commented 7 years ago

@Desmis @Beep6581 Any suggestion on what to do with this patch ? Is the result / trade-off acceptable for you ? It fix artifact on very specific images and make the denoising a little bit stronger on all images (AFAIK). It would be nice to have your though on your own test.

Beep6581 commented 7 years ago

I have not tested this properly yet. Prefer to leave it for 5.2.

heckflosse commented 7 years ago

@Beep6581 Agreed

Desmis commented 7 years ago

Agreed

agriggio commented 6 years ago

Hi, what's the status of this? I think I found another "real-world" test case... :-)

heckflosse commented 6 years ago

@agriggio Alberto, the status didn't change, because it got not enough tests. Can you share your test case?

Beep6581 commented 5 years ago

Still valid?

heckflosse commented 5 years ago

Imho still valid