General profile questions

general questions about Neat Image
Post Reply
overread
Posts: 5
Joined: Tue Mar 10, 2020 6:42 pm

General profile questions

Post by overread »

Greetings.

I've been using NI for a while now, though I've never really delved far into it. As a result in the past I used custom profiles made by others or more recently, I've been using auto generated profiles (most often using an area I pick in the photo) each time I get a new photo. Because most of my photos tend to get shown on the web this has been fairly ok for me since after resizing down for web display, a lot of noise is naturally lost anyway in the transition.

However I'm keen to advance my use of the tool and get better performance out of it.


I've had a read of the profile instructions and I think I've got the idea right, though I want to double check in case I've misunderstood something or missed something out. My plan currently is to

1) Open up the Calibration target in NI on my LCD.

2) Take photos of the screen at different ISO values. As I shoot in RAW it would seem that the only setting I need to worry about is the ISO, since most of the other listed factors that affect the noise are mostly JPEG related. Of course the photos should be slightly blurry in nature.

3) It's not mentioned in the list, but I wonder if for each ISO value if it would be worth taking photos which are not just correctly exposed, but also under exposed and then brightened in editing (to a general default value to represent the same being done to a regular photo). Staggering them to perhaps one and two stops underexposed.
I'm not sure if this is worth doing or if the actual noise patterns remain the same and the under exposure and brightening is just highlighting them more apparently to the viewer.

4) My camera can take photos in 1/3 stop values for ISO and I typically have it set up to use those 1/3rd values. Is it worth creating a fresh profile for each specific ISO value or can you double up and use the nearest ISO value (up or down?) and work with just full stop values of ISO?

5) When in the tool itself do I only need to use the "produce profile with calibration image" tool? Do I need to use any of the fine-tuning command options to further refine the result or is that simply introducing data that will degrade the overall profile

Note I'm aware that shutterspeeds are listed as a potentially contributing factor, however I'm guessing that provided I've decent lighting, I shouldn't need to concern myself with them unless I were having exposures running beyond 30seconds.


Are there any other tips or advice that would be worth taking into account when producing a selection of profiles that I've not noted above? I would be most grateful for any advice/tips regarding this matter.

overread
Posts: 5
Joined: Tue Mar 10, 2020 6:42 pm

Re: General profile questions

Post by overread »

Any neat image users left in the world?!?

lansing
Posts: 59
Joined: Sat Apr 21, 2012 6:52 am

Re: General profile questions

Post by lansing »

Hi I just started playing with Neat Image yesterday but I think I can chip in some thought on this.

To find out whether it's worth to create a different profiles for all the different ISO/exposure for your camera, the best way is to test it yourself. Do it for like different ISO, and then look at the shape of the curve created by the auto profile. If they look very much the same, that mean you can get away with the one profile fit all approach. If the curves looks different, then you'll need to have different profiles for different settings.

The whole point of profiling using a target is to save you time manually tuning it, so don't fine tune yourself, it defeated the whole purpose.

NITeam
Posts: 3135
Joined: Sat Feb 01, 2003 4:43 pm
Contact:

Re: General profile questions

Post by NITeam »

Sorry about the late reply.
overread wrote:
Tue Mar 10, 2020 7:36 pm
2) Take photos of the screen at different ISO values. As I shoot in RAW it would seem that the only setting I need to worry about is the ISO, since most of the other listed factors that affect the noise are mostly JPEG related. Of course the photos should be slightly blurry in nature.
Please remember that any post-processing applied to your RAW image after it left the camera should also be considered and taken into account. For example, if you convert a test image (a shot of the calibration target) from RAW to RGB using your preferred RAW converter working with settings A, then build a noise profile using the resulting RGB image. If you later want to reduce noise in a regular image shot also with the same camera, same camera shooting mode, but converted from RAW to RGB using that RAW converter working with settings B, then the noise in that converted image may be not exactly the same as in that test shot converted with the settings A. So you need to either keep those converter settings constant or keep track of those settings and build separate noise profiles for different conversion settings too. Of course not all conversion settings are influencing noise but some may play a role, for example exposure adjustments, sharpening, noise reduction and similar things that may be offered by the RAW converter.
overread wrote:
Tue Mar 10, 2020 7:36 pm
3) It's not mentioned in the list, but I wonder if for each ISO value if it would be worth taking photos which are not just correctly exposed, but also under exposed and then brightened in editing (to a general default value to represent the same being done to a regular photo). Staggering them to perhaps one and two stops underexposed.
Here we go, different post-processing leads to potentially different noise requiring building separate noise profiles for each of those post-processing scenarios. This complicates things, so usually it is easier to keep the post-processing constant as much as possible.

overread wrote:
Tue Mar 10, 2020 7:36 pm
4) My camera can take photos in 1/3 stop values for ISO and I typically have it set up to use those 1/3rd values. Is it worth creating a fresh profile for each specific ISO value or can you double up and use the nearest ISO value (up or down?) and work with just full stop values of ISO?
I recommend to shoot the test shots in those ISO modes that you actually use for your regular shots that you want to denoise. For example, if you only shoot full stop shots in practice, then you only want to cover those full stops with test shots for profiling. Then the ISO rates of the profiles will exactly match those of the regular shots.
overread wrote:
Tue Mar 10, 2020 7:36 pm
5) When in the tool itself do I only need to use the "produce profile with calibration image" tool? Do I need to use any of the fine-tuning command options to further refine the result or is that simply introducing data that will degrade the overall profile
You would use Auto Profile with Calibration Target when you manually open one of the test shots (of the target) into Neat Image.
You could use the regular Auto Profile on it too, but Auto Profile with Calibration Target is specifically tuned to work with those test shots.
overread wrote:
Tue Mar 10, 2020 7:36 pm
Note I'm aware that shutterspeeds are listed as a potentially contributing factor, however I'm guessing that provided I've decent lighting, I shouldn't need to concern myself with them unless I were having exposures running beyond 30seconds.

Are there any other tips or advice that would be worth taking into account when producing a selection of profiles that I've not noted above? I would be most grateful for any advice/tips regarding this matter.
overread wrote:
Tue Mar 10, 2020 7:36 pm
Note I'm aware that shutterspeeds are listed as a potentially contributing factor, however I'm guessing that provided I've decent lighting, I shouldn't need to concern myself with them unless I were having exposures running beyond 30seconds.
For the first step I would recommend to shoot for different ISOs only. Then if your working shots show some slight variance of noise strength depending on the shutter speed, then I recommend to use Auto Fine-Tune after loading one of profiles built with the test shots to fine-tune it to the current working image. That is the main purpose of Auto Fine-Tune.
overread wrote:
Tue Mar 10, 2020 7:36 pm
Are there any other tips or advice that would be worth taking into account when producing a selection of profiles that I've not noted above? I would be most grateful for any advice/tips regarding this matter.
I would print out the calibration target and shoot it from paper instead of LCD as the screen may give some glares or pixel texture that you do not want to have in those test shots.

Vlad

overread
Posts: 5
Joined: Tue Mar 10, 2020 6:42 pm

Re: General profile questions

Post by overread »

My thanks for you reply!
Sorry it took a while for me to get back, I was trying to get a day to set aside to get some progress on this before reporting back.


One thing I've found that is confusing me with regard to auto-profile with calibration chart is if I don't click on anything on the image itself when I load it into NI and just use the auto calibrate it gives me a chart with a normally pretty high compatibility rating.

However if I select an area in the upper right side (the darker region squares) using the selection tool and then I run the same auto calibrate with profile image, I get a very different histogram result along with a higher (often near 100%) rating. What's odd is that the green channel typically makes a massive drop in this.
Meanwhile if I select an area in the middle (the green block) and do the same I get back to a roughly similar performance as the auto profile with nothing selected.



This is confusing me because its suggesting that the auto-profile is weighting for the green in the middle of the calibration photo; yet if I select elsewhere in the darker regions I can get a higher calibration rating.

I'm unsure which is the better approach, especially as when I pre-select the darker area the colour channels have a roughly similar spread of values (not the same, but the general same pattern); whilst with the middle region the green channel is really spiking high.

NITeam
Posts: 3135
Joined: Sat Feb 01, 2003 4:43 pm
Contact:

Re: General profile questions

Post by NITeam »

Perhaps your images are somewhat unusual. Normally, using the central square as the basis for profiling delivers good results. However if you see better results (both in terms of reported profile quality and actual visual quality of noise reduction in preview), then by all means help Neat Image by manually selecting an area for profiling.

Vlad

overread
Posts: 5
Joined: Tue Mar 10, 2020 6:42 pm

Re: General profile questions

Post by overread »

My thanks Vlad

Out of interest aside from the rating are there any other values that can be used to help judge the quality of a profile. Since comparing noise reduction performance by eye can be somewhat of a tricky thing to achieve when the differences might be very tiny.

I appreciate that chances are I'm splitting hairs on the profiling to some degree.

NITeam
Posts: 3135
Joined: Sat Feb 01, 2003 4:43 pm
Contact:

Re: General profile questions

Post by NITeam »

The quality indication in profiles is an aggregate measure. If you open Profile Viewer (menu Profile > Profile Viewer), you can see some individual measures like Uniformity, Completeness, individual noise levels in different components, presence of clipping. Those can also be used to compare profiles. Ideally you want higher uniformity and completeness, no clipping, noise sample without any details, and green noise levels in frequency components.

Since there are many factors, we have combined them all into a single number that you see as Quality. It is easier to use.

Vlad

overread
Posts: 5
Joined: Tue Mar 10, 2020 6:42 pm

Re: General profile questions

Post by overread »

I compared looking at the dark area selected with the default and interesting whilst the aggregate score was slightly higher on the dark band over the default, the uniformity values were on the whole higher for the default when looking at the detailed information.

Out of interest if you've the time and wish too I've uploaded one of the test photos I've been using

https://i.imgur.com/MuBRVZg.jpg

I'd be interested on your views and on if I've either got something wrong (eg perhaps not blurred enough).

NITeam
Posts: 3135
Joined: Sat Feb 01, 2003 4:43 pm
Contact:

Re: General profile questions

Post by NITeam »

No, the test shot is fine, there is enough blur.

The quality indication is indeed higher with a manually selected area but I expect no difference in the visual quality results with either of the profiles. The quality seems to be reduced for the central area because there are some compression artifacts there that look like details (reduced uniformity) but in fact are not details, so can be safely analyzed as noise.

So I would say that with this image the results will likely be about the same irrespective of the profiling method you choose ('Auto Profile with Calibration Target' or 'Auto Profile' with manually selected area).

Vlad

Post Reply