noisewear artefacts - how to get rid of them?

[kiwisphere]

Hello!

I am new here
I hope I post it in the right branch, was pending btw. general and plug-ins.

I experience some strong artefacts that seem to come out after Imagenomics Noisewear application.
The photograph is challenging, so I am not wondering that it does not work that well, but I hope there is a way to get rid of the artefacts.

Initial condition:
ISO-1600 shot with Nikon D800
1. processed the picture in RawTherapee: exposure compensation +3.6 (I doubt that these are real 3.6, since 0 is darker than the out of cam jpg exposure), lightness +36 and some curves.
2. Imagenomic Noisewear in PS, does a good job, but artefacts remain. (default settings, luminance 70, color 80)
3. Topaz filter. that enhances of course also the artefacts.

Attached some shots.

Now, I understand that a ISO-1600 picutre with all that processing (exp. comp., topaz filter) will look messy. However, it seems the artefacts consist ONLY of black and white pixels (look at the 400x). Hence, I think there is a way to get rid of them.

And, yes, these artefacts are STRONGLY visible even at 20%, esp. because they are all over the picture.

Any ideas?

thanks a lot!

screenshots at 100x:
out of cam jpg:

out fo cam with topaz:

, RAW+3.6 exp. comp.:

, RAW+3.6 after noisewear:

, RAW+3.6 after noisewear and topaz (final picture):


and here in 400x:

RAW+3.6 exp. comp.:

RAW+3.6 after noisewear:

RAW+3.6 after noisewear and topaz (final picture):

 

[IamSam]

Hello and welcome to Photoshop Gurus!

Please forgive, but I'm not sure I understand what your needing. This image is very difficult to determine a. what it is, and b. what are the artifacts as we have nothing to compare it with. It's not like this is a persons face where we would have some idea of what we were comparing.

I would just not use the Topaz adjustment, seems to me that's were it went wrong. Sorry, not much help.

 

[kiwisphere]

The artefacts I mean are the black and white pixels in the last picture. All that salt and pepper.
It seems they initially come from noisewear and get amplified by topaz.
Look at the sequence:
RAW+3.6 exp. comp.: lots of noise
RAW+3.6 after noisewear: noise got reduced, but there are these artefacts now (all that pixels that got not smoothed out)!
RAW+3.6 after noisewear and topaz (final picture): the artefacts got amplified (black/white pixel conglomerates)

Since these are just black, white pixels, I am sure there is a way to get rid of them. It should be quite simple in a code: "Select all pure white and pure black pixels with a certain (small) radius and replace them by the nearest pixel" something like that
I am sure a smart guy at Imagenomic or Adobe implemented that in a tool (maybe even in Noisewear?)

 

[ALB68]

Diffusion filter, set to Darken only?

 

[kiwisphere]

Diffusion filter, set to Darken only?

Thanks ALB68!
I am not sure if I understood it right (do you mean with diffusion filter a Gaussian filter?), but what I did is, I applied a simple Gaussian filter of 0.5 radius. It softens, but at the small radius hardly noticable, exept at 100x - and then it's much better than seeing the artefacts!
I was thinking you mean the diffusion blending mode, but that doesn't make any difference to normal mode.

here the result at 100x



and here the comparison without filter (from previous post) with artefacts visible:

 

[Tom Mann]

I'm not sure exactly what the subject is, but it sure doesn't look like it's going anywhere, so why didn't you shoot at ISO = 200 (or so) with a very long time exposure instead of 1600 and then adding a couple more stops of brightening (ie, the kiss of death) in your raw converter?

Also, do you have the camera set to maximum bit depth raws with either lossless or no compression of the NEF. I ask because these sorts of artifacts almost always come from extreme brightening (ie, +3.6 stops on a compressed image) on a compressed image.

Tom M

 

[kiwisphere]

Tom, I would love to shoot at lower ISO, but that was af f/2.6 and 20 sec. If I shot at 200 instead of 1600, I would have needed a 2 min. exposure. Incl. long exposure noise reduction that would be 4 minutes. Too long if I wanna make a couple of shots in short time. I made during shooting already an exposure bias of +5 as the light source was the fire of a volcano before sunrise that I wanted to get overexposed while the surrounding area should be correctly exposed.

The camera is set to 14 bit and lossless compression (I guess that is fine?)

I am quite happy with the solution of using a 0.5 radius Gaussian filter, to which I came after reading the reply by ALB68, but I am open to better suggestions of course.

Before Gaussian filter (ca. 20% crop)




After Gaussian filter (ca. 20% crop)

 

[Tom Mann]

OK, I now have a better appreciation of what you were trying to capture. You may already know this, but FWIW, the best noise reduction is always done on the raw data before it's demosaiced, ie, on a sensor site by sensor site basis instead of a pixel by pixel level such as is done when one applies NR later in the workflow, eg with Noiseware.

I presume you have experimented with the NR options in Raw Therapee. If not, I would give it a try. It's always much better to nip the production of those artifacts before they are even formed, rather than trying to remove the artifacts after the fact.

Also, the general feeling is that the NR for Nikon DSLRs in the latest versions of Capture NX2 and ACR are significantly better than the NR algorithms in other raw converters, so, if you haven't already tried these, you might want to give these two a try. Also, last year's upgrade to DxO's NR (high priced) is reputed to be significantly better than even that of ACR and Capture NX2, but that's only from a small population of users and not the most rigorous of testing.

Finally, a question: Are these pix for artistic or scientific purposes? If for scientific purposes (ie, where non-realistic colors are acceptable), you might want to consider image intensifiers and mid_IR sensors. Unfortunately, I don't think removing the IR cut filter from a normal DSLR will do much good as it will only get you into the near IR.

HTH,

Tom

 

[Tom Mann]

Also, if you can return to that location and money is no object, the relatively new D750 is reputed to have significantly better High ISO / noise performance than even the new D810. See:
http://nikonrumors.com/2014/09/14/e...ow-about-the-new-nikon-d750-dslr-camera.aspx/


Tom M

 

[Tom Mann]

And if none of the previous thoughts (ie, different exposure, different raw converter, different camera) are useful, here's about the best I could do using only PS's native tools. If you are interested, let me know and I'll write it up. It's quite simple.

Tom M

 

[kiwisphere]

And if none of the previous thoughts (ie, different exposure, different raw converter, different camera) are useful, here's about the best I could do using only PS's native tools. If you are interested, let me know and I'll write it up. It's quite simple.

Tom M

Hi Tom!

That looks great! It is exactly what I was looking for, i.e. get rid of the black and white pixels!
I would highly appreciate if you could tell me how you processed it!

thanks!

 

[Tom Mann]

I'm glad it seems to be what you were looking for. Unfortunately, I'm just heading out for the evening, but I'll write it up late tonight when we get back.

Tom

 

[kiwisphere]

Great!
Looking forward!
Enjoy your evening

 

[Tom Mann]

Hi Kiwi -- I'm back for a few minutes, but then going out again. Anyway, since my last post, I realized that I had already described this technique on this forum:
https://www.photoshopgurus.com/foru...ation-girl-post1533634383.html#post1533634383

For your photo, I think I used a radius of around 6 px for the D&S filter on one layer, and around 2 or 3 px for the other layer with threshold settings of around 30-50%, but adjust to taste.

If you have any questions, don't hesitate to ask.

HTH. More later ...

Tom