Sure, I'm not pointing the finger at profiled denoise specifically as my
comments apply equally to any of the other tools, and I agree the profiled
denoise is a real good starting point for processing images in general.
I merely draw the observation that when shadows are boosted it seems as if the
required denoise in those areas seems fairly equivalent to an image taken at
the higher boosted ISO values. Obviously if I have chosen a simple exposure
increase across the image that's easy to compensate for, if it is S&H or
similar boosting shadow exposure by some unknown internal value on selective
areas it's rather harder to come up with reasonably optimal denoise levels. It
would be great if it were possible to apply that sort of localised ISO denoise
optimisation automatically - but if it isn't practical to do that I'll shut up.
Rgds,
Rob.
From: johannes hanika [mailto:[email protected]]
Sent: 16 December 2013 17:51
To: darktable-devel
Subject: Re: [darktable-devel] white dots hight ISO
hey,
profiled denoise is purely driven by numbers.. the input is measured noise
variances and the output is adjusted to minimise some mean square error if you
don't do any other processing. it does that by minimising the risk of
destroying data that was actually present in the input, while trying to remove
as much noise as it dares to. that might not be what you want or what looks
most pleasing to your individual taste. but i would argue it's still the best
starting point for further processing, even if modules you apply on top will
change the perception you have of the noise in the image.
also there are quite a couple of cameras for which the noise model doesn't
really match reality (some of it due to secret in-camera processing, some
inherent to the poissonian/gaussian prior) so there is some sub-optimal noise
removal in darker regions. if you brighten that it'll look even worse, of
course.
doing it the other way around (remove noise at the end, do processing in
between) makes very little sense to me. first off, modules will be driven by
noisy input and produce weird artifacts because of that. secondly, you don't
only scale the values, but you will end up with unpredictable noise
characteristics with interdependencies between pixels and non-trivial
distributions (probably both the natural image prior and the noise model will
be off).
-j.
On Mon, Dec 16, 2013 at 6:30 PM, Rob Z. Smith
<[email protected]<mailto:[email protected]>> wrote:
Thanks Jo, I wonder if I have misunderstood.
Say I take an image at 800 ISO and profiled denoise handles it just fine, but
then I come along and hit it with shadows and highlights or otherwise boost
shadows by two stops to pull out detail. My eyes now tell me that I can see
noise in those lighter shadows they couldn't see before and the denoise
settings needed to deal with those pushed shadows are sort of equivalent to two
stops higher at 3200 ISO not 800. Is this my imagination or is that how it
actually works?
Rgds,
Rob.
From: johannes hanika [mailto:[email protected]<mailto:[email protected]>]
Sent: 16 December 2013 15:39
To: Rob Z. Smith
Cc: Halgeir Kjønås Rennehvammen; darktable-devel
Subject: Re: [darktable-devel] white dots hight ISO
On Mon, Dec 16, 2013 at 2:03 PM, Rob Z. Smith
<[email protected]<mailto:[email protected]>> wrote:
While we are on the subject of denoise - if I could choose something for my
Xmas list it would be for dt to keep track of how much each pixel (or area) of
the screen has been boosted in brightness by the various modules and then offer
denoise (or a mask) based on the effective ISO we have raised those pixels to.
that's why noise reduction is done before all the modules mess with your data.
-jo
This email is confidential and is intended for the addressee only. If you are
not the addressee, please delete the email and do not use it in any way. Please
note that any views or opinions presented in this email are solely those of the
author and do not necessarily represent those of the company. NHBC reserves the
right to monitor all email communications. The recipient should check this
email and any attachments for the presence of viruses. The company accepts no
liability for any damage caused by any virus transmitted by this email. NHBC,
the National House-Building Council, is limited by guarantee in England, No
320784. Registered address: NHBC House, Davy Avenue, Knowlhill, Milton Keynes
MK5 8FP. NHBC is authorised by the Prudential Regulation Authority and
regulated by the Financial Conduct Authority and Prudential Regulation
Authority. NHBC Building Control Services Ltd, registered by guarantee in
England with Company No. 01952969. Registered address: NHBC House, Davy Avenue,
Knowlhill, Milton Keynes MK5 8FP. NHBC Services Ltd registered by guarantee in
England, No 03067703. Registered address: NHBC House, Davy Avenue, Knowlhill,
Milton Keynes MK5 8FP. If you make a claim under a Buildmark policy your
personal details will be stored and processed in accordance with the Data
Protection Act. Your personal details may be passed to others involved with
your claim such as the original builder, or a consultant or remedial works
contractor that we may employ in connection with your claim(s) and matter
ancillary to your claim(s). Other than disclosure provided for in this
statement, we will not pass any data about you to any other party without your
permission unless we are required to do so by law.
------------------------------------------------------------------------------
Rapidly troubleshoot problems before they affect your business. Most IT
organizations don't have a clear picture of how application performance
affects their revenue. With AppDynamics, you get 100% visibility into your
Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro!
http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk
_______________________________________________
darktable-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/darktable-devel
