[Leica] comparison of 4 noise reduction programs

Don Dory don.dory at gmail.com
Wed Apr 19 16:45:59 PDT 2023


Howard, the algorithm in Topaz frequently has an issue with letters.   As
to DxO, I was at the middle level of effect with two higher levels
available.   I chose not to go there as I thought an intermediate level
would be a better comparison.   I found that if I mess with Topaz I can get
a good result but that involved fueling with the masks Abbott balancing
strength and clarity.

I purposely picked a very poor image to show what the extreme would be.   I
can say that on images shot around 12000 ISO with some contrast in the face
DxO dies a better job than the Adobe product.   It's why I have all three.


Others have commented on the stand alone Topaz products,  I find their
DeNoise works very well on moderate noise if you don't need sharpening: I
will choose it when speed is important.

On Wed, Apr 19, 2023, 5:59 PM Howard L Ritter Jr via LUG <
lug at leica-users.org> wrote:

> Don, the stunning superiority of the LR Enhance image in terms of
> sharpness of the subject’s hair, face, sweater, and nametag (or
> alternatively, the comparative lousiness of the other results) beggars
> belief. I have to wonder whether something didn’t go wrong. Among the
> three, only in the TopazAI is her face in focus, while it’s not even close
> in the other two. Something other than noise level is involved in the
> differences in resolution and focus. I can’t imagine there being any way
> that the level of detail in the LR image could have been recovered from a
> primary image as out of focus as the other two would suggest, even with a
> dedicated sharpening program, let alone merely a noise-reduction program.
>
> For example, in the LR image, look at the leftmost strand of hair, which
> goes from highlighted to darkly silhouetted against the background figure
> as it sweeps upwards. In the other images, there is not even the merest
> suggestion of this latter part, only the uniform blur of the background.
>
> Look at her nametag and ribbon. In three images, including Topaz AI, the
> printing isn’t even recognizable as such. But in the LR image, her name is
> not only recognizable, not only easily readable, but actually sharp, even
> down to the presence of the demarcation that shows her last name is
> ‘Martin’ and not ‘Martln’. And it’s puzzling that the TopazAI image, which
> has the face in so much better focus than the Topaz DeNoise and the DxO,
> doesn’t do any better on the nametag.
>
> How can this be? Did one program perform magic, or did the other three
> actually degrade details?
>
> Can you post the unprocessed image for comparison? That would be extremely
> interesting.
>
> —howard
>
> > On Apr 19, 2023, at 6:50 AM, Don Dory via LUG <lug at leica-users.org>
> wrote:
> >
> > Greetings to all.  The attached files are the same image processed in 4
> > noise reduction programs.  The base image was shot at 25600 ISO and
> cropped
> > to less than 50% of the image.  The camera was a Sony A1 and the lens
> was a
> > Sigma 105 F1.4 at 1.4.  Take a look and decide, the only processing was
> the
> > noise reduction, cropping, and lightening the overall image.
> >
> > First is LR enhance:
> >
> >
> http://gallery.leica-users.org/v/don_dory_gmail_com/Noise/LR+enhance.jpg.html
> >
> > Next is the older Topaz denoise:
> >
> >
> http://gallery.leica-users.org/v/don_dory_gmail_com/Noise/Topaz+denoise.jpg.html
> >
> > Topaz AI:
> >
> >
> http://gallery.leica-users.org/v/don_dory_gmail_com/Noise/Topaz+AI_.jpg.html
> >
> > DxO deep prime although not the strongest selection:
> >
> >
> http://gallery.leica-users.org/v/don_dory_gmail_com/Noise/DxO+deep+prime.jpg.html
> >
> >
> >
> > --
> > Don
> > don.dory at gmail.com
> >
> > _______________________________________________
> > Leica Users Group.
> > See http://leica-users.org/mailman/listinfo/lug for more information
> > .
>
>
> _______________________________________________
> Leica Users Group.
> See http://leica-users.org/mailman/listinfo/lug for more information


More information about the LUG mailing list