Tuesday, December 26, 2023

An Example Of High ISO (12,800) Noise Reduction Of A 26mp Fujifilm X-H2S File And A 20mp OM-1 File (UPDATED)

Original 20mp uncropped image shot at ISO 12,800 with the OM-1. (click to enlarge)

Original 26mp uncropped image shot at ISO 12,800 with the Fujifilm X-H2S. (click to enlarge)

Recently, I wrote the post comparing the OMD Systems OM-1 with the 100-400mm lens and the Fujifilm X-H2S with the 150-600mm lens for bird photography, which you can read here, I mentioned I saw a difference in detail in very high ISO files between the two cameras after noise reduction was applied.  This is just a short follow-up of that statement with an example.  I thought some of you may find this interesting.

I was drawn to this scene by the light.  It was early on a December morning, just 30 minutes after sunrise, and a shaft of light was partially lighting this Great Blue Heron which was also trying to stay somewhat secreted under the overhanging branches.  To my eye, this scene was something I found attractive.  When I saw that I would be shooting at ISO 12,800, that did not deter me as I've used such high ISOs in the past (ys, even with micro4/3!) and knew that, with the advanced noise reduction plug-ins now available, what was prohibitive in the past is now possible.

The image at the top of the screen is the original uncropped image shot at ISO 12,800 by the OM-1 with the 100-400mm f/4.5-6.3 lens.  The one just below is the original image shot at ISO 12,800 with the Fujifilm X-H2S and the 150-600mm f/5.6-8 lens.  Both have extreme digital noise.  The images are very similar but a little different as they were made within a couple of minutes of each other.  Both lenses were at maximum aperture and both files are raw files.

Below are two approximately 100% crops of the above images showing the excessive noise.

Approximately a 100% crop of the top image shot with the OM-1 to show extent of noise.
(click to enlarge)

For comparison this shows the same but with the X-H2S. (click to enlarge)

Below are the images in their entirety after being run through DXO's Deep Prime HD noise reduction software.  

Original uncropped OM-1 image after being treated with DXO Deep Prime HD noise reduction.
 (click to enlarge)

For comparison the original uncropped X-H2S image, again, after being treated with DXO Deep
Prime noise reduction. (click to enlarge)

Now, below are approximately the same crops of the two images that were treated in DXO’s Deep Prime HD noise reduction.

Again, an approximately 100% crop of the OM-1 image that was treated with the DXO
noise reduction. (click to enlarge)

Again, the same of the X-H2S image treated with the DXO noise reduction. (click to enlarge)

As best you can, considering the compression and whatever else Blogger does to the image files, I'll let you judge for yourself as to which looks better after the treatment and if either are usable in your judgment.  In either case, the removal of such extremes noise is remarkable.

To me, when viewing them on my monitor, the OM file has more detail and less of a plastic look in some places than does the Fujifilm file.  This is a sample of one.  Don’t draw any far reaching conclusions from this one example.  That said, it is typical of what I saw on this particular bird photography outing.

This is my final edited and cropped version of the OM-1 image file.  Like it?  Don't like it? 
What do you see wrong with it? Personally, for a micro4/3 camera and a not too expensive lens
shot at ISO12,800, I find the image quite remarkable! (click to enlarge)
OM-1; 100-400mm f/4.5-6.3 lens @ 276mm (553mm 35mm equiv.); 1/1600th sec. @ f/6.2

UPDATE DECEMBER 27, 2023   

A commenter wondered how the files would look if I had run the them through Topaz Photo AI to see how they compare to the ones posted here.  Below are the two files after running through that program.  I did not alter the "suggested" settings by the Photo AI as DXO doesn't provide any changes to its settings.

This is the OM-1 uncropped file after being processed in Topaz Photo AI at its suggested settings.
The raw file was sent to Photo AI in the "Plug-In Extras" menu, not a converted TIFF file using
the "Edit in" menu item.  It looks almost identical to the DXO file.  Almost.  I think the DXO file
just edges it out by a hair, so to speak, in detail and noise reduction. But that is totally subjective
on my part.  You may like the Photo AI file better.  (click to enlarge)

This is a cropped file of the above image.  You may not be able to see what I see when I say
the DXO version looks better to me.  As I said, they are very close.  Looking at the two processed
files casually, you probably won't notice a difference. (click to enlarge)

This is the X-H2S file sent through Photo AI exactly as above.  The differences between the OM
files and the Fujifilm files is much more obvious here.  As you will see in the approximate 100%
crop below, there are purple colored artifacts, more visible noise and still a tad less detail in the
X-H2S file treated in Photo AI.  (click to enlarge)

Again, you may not be able to see what I am seeing after Blogger treats the files with compression
and whatever other things they do, but in the original on my monitor there are clear purple artifacts
in the feathers on the wings and elsewhere.  I believe it is Choma (color) noise that Photo AI did not 
remove. It seems to me that has been an issue in the past.  It is best to remove the color noise in 
Lightroom before sending the file to Photo AI.  But, then again, you can't do that with a raw file, 
only with a converted TIFF using the "Edit in" menu item, which is a bit of a disadvantage.
 (click to enlarge)

Based upon what I see in these additional files treated with Photo AI, for now, DXO Deep Prime HD would be my first choice for extreme noise reduction.  That said, Photo AI is continually improved and, with some additional adjustments, most likely could match the DXO results.  The nice thing about DXO is that there are no adjustments.  Much more expedient in not having to 'mess' with a file.  But....in other files, that may be a negative.  Remember, this example is a sample of one—one composition, one lighting situation.  Don't draw final conclusions based upon a single example.  That is not prudent.

Join me over at my website, https://www.dennismook.com 

Thanks for looking. Enjoy!  

Dennis A. Mook  

All content on this blog is © 2013-2024 Dennis A. Mook. All Rights Reserved. Feel free to point to this blog from your website with full attribution. Permission may be granted for commercial use. Please contact Mr. Mook to discuss permission to reproduce the blog posts and/or images.

6 comments:

  1. Very fine wildlife image. Direction of the lighting worked out great with the position assumed by the heron. For sure, the DXO noiseware did the job.

    In my experience, noiseware would still have been necessary in these conditions had a full frame camera been used.

    I am curious if the Topaz Denoise would have been as effective.

    ReplyDelete
    Replies
    1. Thank you. As you saw, I did not run the two files through the Topaz software. I guess I should have but I was trying to simplify the post. However, tomorrow I’ll run the two files through Topaz Photo AI at the software’s “suggested” settings and add the results to this post. ~Dennis

      Delete
  2. Awesome picture, I like it too. It is amazing what photography software can do with noise from high ISO's. All the best in the New Year.

    ReplyDelete
    Replies
    1. Thank you on both accounts. I wish you a Happy New Year as well! ~Dennis

      Delete
  3. Excellent shots - what’s truly remarkable to me is how “competitive” the MFT sensor appears to be against all odds.
    Not fully sure how NR in this case works but I presume it applies further averaging interpolation layer atop the RGB values to even them out. Would the config of the x trans RGB pixels be the main reason for the plasticky look following such extreme demonise?

    ReplyDelete
    Replies
    1. Thank you. I’m guessing, as you have, as to why the difference. It may, indeed, be the difference in the OM using a Bayer sensor and the X-H2S using an X-Trans sensor. Most software does at least a little better with the predominant Bayer sensor, it seems. The main three arguments against micro4/3 is a) digital noise, b) no shallow depth of field and c) less dynamic range. That said, you can see here that digital noise can be neutralized. Lightroom’s new Lens Blur feature is amazing in blurring the background and foreground if less depth of field is wanted and most people don’t realize that the differences in dynamic range are reduced dramatically once you start raising the ISO above base. The reduction of dynamic range is almost a direct inverse relationship with raising the ISO. In many cameras at ISO 3200 or so, dynamic range is cut half and the differences between the dynamic range in different sensors is reduced as well. ~Dennis

      Delete