Luminous Landscape Forum

Raw & Post Processing, Printing => Digital Image Processing => Topic started by: Garnick on January 10, 2020, 12:52:38 pm

Title: Best Noise Reduction-Once Again
Post by: Garnick on January 10, 2020, 12:52:38 pm
Although this topic has been approached a number of times, as I have just read in a search, I would like to bring it up to present times and applications, since there seems to be a number of ways to take care of noise in a digital image.  Therefore I'm hoping to see some replies that take on this issue in a deep and yet easily understandable manner.  I realize that obviously there will be various opinions, but I believe the LuLa members will take this issue as something of importance and reply with appropriate reasons for their individual opinions etc. 

Thank you to ALL  :)   
Title: Re: Best Noise Reduction-Once Again
Post by: leuallen on January 10, 2020, 05:40:53 pm
I use this method for B/W, m4/3 at iso's 6400-8000. Subject matter matters: I am processing a performer in a bar at very low light levels. I have a background in film with Tri-X (60's-70's+) and look for similar grain structures. I want some grain for the authenticity, mood, and grittiness it imparts. But the iso's I have to to use are way to grainy.

I use Topaz Denoise AI. It is very slow because of all the processing it has to do. I convert my files to DNG and send them to Photoshop with no Lightroom processing. I make a copy layer which is used for the denoise operation. I start with the Auto setting and work up and down from there. The goal is to get a completely grain free as much as possible. This will look terrible and plastic. Process and look at the denoised layer. Be sure to check the entire image as there may be spot areas that do not look too good and are still grainy. Undo and redo the denoise process until you get a clean image.

Now modify the layers opacity setting and check the grain at layer opacity 100% until you get a pattern in the background that you like. Check different areas as the grain will vary over the image. If it looks good all over, you are done. I set the layer base opacity so that the background has a very fine grain structure. This works well for the type of image I am processing. If you are processing a different type of image, a landscape with sky for example, you may set you layer base opacity for the sky.

The advantage to this method is that if you try to get the perfect grain in Topaz you will eat up a lot of time as each trial requires reprocessing. It is very quick to effect changes by altering the opacity till you get what you want. Once you get a denoise setting it can be used on the following images in a set. The following images are much quicker because I do not bother to process a preview: open the file in Denoise and immediately process.

It us unusual for me to get a good overall result as above. If I get the background is looking good, the performers skin usually looks waxy and over denoised. Details in the performers clothes can usually stand more grain. So I expand the technique as follows: I add a reveal all mask (white). The important point at this stage is that you can further change the grain by operating on the mask. If you paint on the mask with a darker grey, the grain will increase, more of the grainy background layer will show through. Use a lighter grey or white and you will get less grain. Now using a soft brush set the brush color to some grey and use a low opacity or flow. You can increase/decrease the grain by painting on the mask. Values of grey and opacity/flow are determined by experiment but once you have values they can be used for most of the following images. For this set of image I used 94% for the layer, 60% white for the brush, and flow off 1-10%

As I want to bring out more grain on the performer I set the grey to something like 60% or 70% (darker than the layer opacity set above) and the flow to 1-10%. Then with the image at 100% I start painting on the mask with the image visible. I go around the outlines testing an area and adjusting the flow/opacity till I get a result I like. Then I finish the outline and fill in the central areas. Check the image for areas that need more work. Add noise with with more brush strokes. Reduce noise with an eraser at low opacity 1-10%.

The file is saved with layers for further manipulation and sent back to Lightroom. In my example, the file in Photoshop is in color because the file is not altered by Lightroom. On its trip back to Lightroom it is converted to black and white, split toning, exposure adj. and so on to finalize the image.

After processing in Lightroom if the grain is not right the original is sent back to Photoshop to have the grain adjusted. This may occur because the original was underexposed or for other reasons. You do not want to adjust the exposure before sending to Photoshop because Denoise works best on an unadjusted file. You may get away with it if the noise is not too bad but I am working on very noisy images so I follow this practice.

Back in Photoshop. If the file is overall a little too noisy, increase the opacity of the denoised layer. For example, my denoised layer is usually around 90-94% so I increase it close to 100%. If the background is OK but the subject (area that has darker opacity on the mask) is to noisy, I lighten these areas by using a curves layer on the mask. This is not an adjustment layer, it modifies the pixels directly. Select the mask and then ctrl-m. Move the curve upward in the highlight area to lighten the curve values and reduce noise. Downward will darken and increase noise. This is effecting the painted areas of mask without changing the background density.

First three example at 100%. Last is final Lightroom processed file.
Title: Re: Best Noise Reduction-Once Again
Post by: JaapD on January 13, 2020, 02:06:32 am
I have experience with Topaz Denoise AI but have mixed feelings. In some cases the noise reduction seems ok, but the results are never the same over a range of similar images. The reproducibility is low. Even worse is the fact that in some areas of the image the noise reduction kicks in relatively hard and in other areas not, giving a very unnatural result.
I follow the incremental (so far small) improvements on the AI software packages but up till now I donít really use it on the final images.

I have heard of good results by the DXO noise reduction algorithm, although I have no experience on this myself.

My all-time favorite is NeatImage. What I particularly like here (wouldnít want anything else...) is that the program actually measures the noise spectrum and then mathematically creates the correction filter. This is contrary to many noise reduction programs where you need to perform a visual check yourself and move the sliders until youíre kind of happy with the result. Reproducibility is high.

Title: Re: Best Noise Reduction-Once Again
Post by: Martin Kristiansen on January 13, 2020, 05:58:54 am
I also use Neat Image as my most trusted noise reduction software. I have downloaded some canned profiles from them for my cameras as with some images it can be tough to find a clean area to profile. I sometimes also use LR, itís actually quite good but takes a bit of playing around. I also use Topaz AI but it is very hit and miss and I agree that it can be aggressive in some areas and not strong in others. When it works itís great but more often than not it doesnít work for me. I keep updating and checking on Topaz, it is improving and perhaps in the future it will be a real thing.