On Thu, Nov 3, 2016 at 6:59 PM, KOVÁCS István <k...@kovacs-telekes.org> wrote:
> The HSV-based solutions look inferior to me. Last September, I started
> a thread, but it did not get far. See the message quoted below - the
> sample images are available at
> http://tech.kovacs-telekes.org/files/dt_profiled_denoise_colour_vs_hsv_colour/
>
> Thanks,
> Kofa
>
>
>
> ---------- Forwarded message ----------
> From: Stéphane Gourichon <stephane_darkta...@gourichon.org>
> Date: 12 September 2015 at 09:19
> Subject: Re: [Darktable-users] profiled denoise - colour vs HSV colour
> To: darktable-us...@lists.sourceforge.net
>
>
> Le 11/09/2015 22:44, KOVÁCS István a écrit :
>
> Even when fit to the screen (1920x1080), I find the plain
> 'colour'-blended version (DSC_8834.jpg) less noisy *and* sharper than
> the HSV-colour-blended one (DSC_8834_01.jpg). Zoomed in, the noise
> seems more 'patterned' on the HSV-blended version (shows some kinds of
> structure, patches, homogeneous areas with more defined, thicker
> borders).
>
>
> The better noise characteristic you mention (better on DSC_8834.jpg
> than on DSC_8834_01.jpg) seems pretty clear to me.
>
> From my experience, the pattern you mention have similar size as
> typical *color* noise patterns in photographs.
>
> I'm not a human vision specialist yet have some past experience during
> my Ph.D. in robotic vision where the colorspace used for computation
> has sometimes important impact on the effectiveness/robustness of some
> algorithms.
>
> In the case of your observation, I would guess it is related to the
> fact that performing image operations on HSV colorspace misfits/mixes
> what human vision perceives as luma and chroma. As a consequence, the
> noise reduction performed on S and V "bleeds" into what the eye
> perceive as luminance, where it appears as the patterned noise you
> mention.  This theory goes in the same direction as your observation.
>
> For more information, read the paragraphs above and below the turtle
> photographs on https://en.wikipedia.org/wiki/HSL_and_HSV#Disadvantages
>
> Can anyone elaborate the difference between "color" and "chroma"
> blending mode in this respect ?
This may have been pointed out already, but
https://github.com/darktable-org/darktable/blob/master/src/develop/blend.c#L2035-L2036
_blend_color(): color blend; blend hue and chroma, but not lightness -
blend hue along shortest distance on color circle

https://github.com/darktable-org/darktable/blob/master/src/develop/blend.c#L1898-L1899
_blend_chroma(): chroma blend - just blends C channel in LCH (for Lab
based iops), or S channel in HSL (for RGB based iops)

> Regards,
>
> --
> Stéphane Gourichon
Roman.

> ------------------------------------------------------------------------------
>
> _______________________________________________
> Darktable-users mailing list
> darktable-us...@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/darktable-users
>
> On 3 November 2016 at 16:18, Roman Lebedev <lebedev...@gmail.com> wrote:
>> On Thu, Nov 3, 2016 at 5:57 PM, Andy <t3k...@gmail.com> wrote:
>>> Hi all.
>>> I'm new to darktable and somewhat new to raw image processing so please
>>> excuse me if this is a naive question.
>>> I was trying to denoise some images from my Sony a7 II R shot at 51200 ISO.
>>> I followed some tutorials that recommended applying wavelet denoise with
>>> blend=color and non-local means with blend=lightness.
>> Also do try blend=hsv color and blend=hsv lightness.
>>
>>> I noticed some bright green highlights appearing. I played around with the
>>> demosaic-ing and noticed that it even happened with
>>> the "rough monochrome" demoisaicing and so I thought it might be worth
>>> asking if this is a bug or some odd interaction of the filters.
>>> You can find a picture here:
>>> https://whttps://www.dropbox.com/s/ozjy3dk84zgvs6s/DSC00961.ARW?dl=0 and the
>>> xmp here: ww.dropbox.com/s/fr24nz7zgz7xgd9/DSC00961.ARW.xmp?dl=0
>>>
>>> If you go to the highest zoom level with wavelet denoising on the color and
>>
>>> raw monochrome demosaic, you will see the bright green artifacts.
>> Monochrome demosaic is NOT meant to be used to "monochromly" demosaic
>> the normal raw files. It is only meant to be used for the raw files from 
>> cameras
>> with color filter array physically removed, scraped off.
>>
>>> They disappear when disabling the color denoising with wavelets.
>> The profiled denoise is tuned to the properly-demosaiced image.
>> If it does exhibit some artifacts when used on non-demosaiced input
>> (because that is basically what monochrome demosaic does),
>> i would not be surprised.
>> Because monochrome is, as far i'm concerned, the exact opposite of color.
>>
>>> Using any other demosaic algorithm, you'll see magenta artifacts in the
>>> highlights as soon as you turn the color denoising on.
>>
>>> My question is: is this expected? In particular with the monochrome demosaic
>>> the artifacts seem very odd.
>> I'd *personally* say yes.
>>
>>> But in general
>>> I'd love to get rid of the bright green / magenta artefacts in the
>>> highlights when doing the color denoise.
>>>
>>> Thank you!
>>>
>>> Andy
>> Roman.
>>
>>> ____________________________________________________________________________
>>> darktable user mailing list
>>> to unsubscribe send a mail to darktable-user+unsubscr...@lists.darktable.org
>>>
>> ____________________________________________________________________________
>> darktable user mailing list
>> to unsubscribe send a mail to darktable-user+unsubscr...@lists.darktable.org
>>
____________________________________________________________________________
darktable user mailing list
to unsubscribe send a mail to darktable-user+unsubscr...@lists.darktable.org

Reply via email to