New version 7.0 lower quality profiles

questions about practical use of Neat Image
Post Reply
franicma
Posts: 6
Joined: Sun Aug 21, 2011 12:45 pm

New version 7.0 lower quality profiles

Post by franicma »

I have performed several test on scanned film (Coolscan 9000) and the new version 7.0 produces materially inferior results than version 6.1.

This is most noticeably on people faces or fine textures in clothing where version 6.1 does an excellent job retaining detail but version 7.0 produces a smeared output (all filter settings identical).
franicma
Posts: 6
Joined: Sun Aug 21, 2011 12:45 pm

Example

Post by franicma »

This is an example of Kodak Portra 400 processed with both versions of Neatimage.

http://shutterclick.smugmug.com/Photogr ... nMwwx-O-LB

I hope this problem in the algorithm to generate the profile gets fixed soon.
franicma
Posts: 6
Joined: Sun Aug 21, 2011 12:45 pm

Post by franicma »

Here is the full scan for those interested to test it:

http://shutterclick.smugmug.com/Photogr ... cGpJF-O-LB
NITeam
Posts: 3173
Joined: Sat Feb 01, 2003 4:43 pm
Contact:

Post by NITeam »

Please make sure you use the same area for noise profiling. The difference in the noise area is likely to be the cause of the difference in the output.

Neat Image tries to find a good area for analysis but of course it is not always perfect in that choice, so you have to control what is selects.

Vlad
franicma
Posts: 6
Joined: Sun Aug 21, 2011 12:45 pm

Post by franicma »

Thank you for responding Vlad. I tried setting the same sample manually but the algorithm to generate the profile seems to have changed. You can try running it on the full scan (posted above).
franicma
Posts: 6
Joined: Sun Aug 21, 2011 12:45 pm

Post by franicma »

Now if I generate the profile with version 6.1 and load it with 7.0, then 7.0 works correctly. It is just the profile generation that was affected.
NITeam
Posts: 3173
Joined: Sat Feb 01, 2003 4:43 pm
Contact:

Post by NITeam »

The profile generated by v6.x from this image in automatic mode underestimates the noise levels by about 10-15%. Which, as a lucky side-effect, makes more fine details to be preserved, but of course reduces less noise too.

The profile generated by v7 from the same image contains a more precise measurement, which allows to reduce more yet some fine details may be affected too.

The second case is better, because the measurement is more accurate, which is a better starting point than a lucky coincidence of the first case.

Also, for a test, I have built two profiles, one in v6.1 and one in v7.0, using the same area on the cheek of the girl (one of the better choices as compared with the very dark background) and the resulting profiles are quite close, they both show about the same overall noise level, not that 10-15% difference that you see in fully automatic mode. So, if you build a noise profile accurately, the result is the same as in version 7 used in auto mode. If you just rely on automatic Auto Profile in v6.x then it may be a bit inaccurate in both directions (underestiming or overestimating the noise). Using a more accurate noise profile from v7 is a better starting point. The next step would perhaps be to reduce the noise reduction amount in the luminance channel a bit, to better preserve those fine details, while the chrominance channel are filtered at full strength, since that color noise is the most distractive.

Vlad
franicma
Posts: 6
Joined: Sun Aug 21, 2011 12:45 pm

Post by franicma »

Thank you for the explanation and the time to test it.

I performed several tests on different films scanned with the Coolscan 9000 and it seems like version 6.1 preserves more detail while reducing noise similarly. This could be just for the Coolscan since it captures more detail than flatbed; whereas 7.0's new algorithm may be better for flatbeds.

I will play with the settings as you recommend to try to match the results from 6.1 with 7.0.
Post Reply