Intelligent USM

[Paul Escudier]

The only feature that needs adding to eclipse all other scanning software is an intelligent USM featurethat adjusts sharpening according to enlargement factor, picture content e.t.c.
Basically the software would analyse the picture and taking into account various criteria estimate what the acceptable sharpness levels should be, the user can then make fine adjustments if necessary. A similar concept has been on the scitex scanners for years, and the other day I saw a photoshop plug-in called (I think!) Nik Sharpener Pro that did exactly this. If the software could just change USM settings according to enlargement factor, just to get it in the right ball park for minor adjustment, that would be a huge move forward.

[President_LSI]

Intelligent USM (Unsharp Masking)

We have in fact such a function within SilverFast Ai called "Auto-Sharpen". Still there is room for further development and we have not fully implemented that function for all scanners.

This function also has selections like "Less Auto-Sharpen" or More Auto-Sharpen".

[A93]

I have been UNLUCKY enough to try Nik Sharpener Pro and can tell you that the results you get are terrible if you want to sharpen there is no better way than USM and tweaking the image till its right.

[Paul Escudier]

A93,

Ok,I've just been playing with the demo of Nik Sharpener, you're right it's terrible, nice idea though. What I'm talking about I suppose is according to enlargement factor and film format that the USM control gives you a better starting point each time rather than the standard default so that a very quick tweak will give top quality results. Even a selection of presets, setting matrix size and intensity according to enlargement factor would help new users. As for the Autosharpen function I've got to say it's not very good at all, and produces some very strange results

[President_LSI]

Enhanced SilverFast USM function

You can expect that we will invest more development in the near future into Unsharp Masking. Any good ideas and proposals are very wellcome. We want to enhance our function so any realistic expectation can be fulfilled.

[poliwog]

Hi!

Don't think this can be done (or should be) in SF or any scanner software, but the best solution to sharpening is to create an edge mask prior to applying USM. There are a number pf PS Actions which address this, and I believe their complexity is far too great to be executed in a scanner package.

By the way, Nik Sharpener is indeed HORRIBLE! It oversharpens everything to hell.

Les

[Carson Wilson]

Has the sharpening feature been improved any since August?

This issue has become very important to me since I discovered that sharpening and scaling are closely related. A severely "oversharpened" scan loses all of its artifacts as soon as I resample it down to a smaller size.

I also notice that resampling my SilverFast output to a smaller sizes greatly reduces jaggedness of angular edges in the image. So for example, if I scan a negative at 1350 dpi and then resize it to 50% of its size, jaggedness is noticably reduced when compared to a scan taken at 625 dpi, even though both images are the same size.

Why? I think this results from a combination of reduced "grain aliasing" due to the higher resolution initial scan (there are web pages about this I can reference here if anyone's interested), and from antialiasing performed by the image program when the image is resampled to a smaller size.

So, what I would like SilverFast to do is:

• Perform the initial scan at as high a resolution as possible;
• Apply its unsharp mask at this high resolution;
• Once the above is complete, resample down to my desired resolution.

As suggested above, knowing the desired final resolution could allow SilverFast to "tune" the amount of USM applied to the original, higher-resolution data. In general it appears that the greater the downward resample scale, the more USM we can get away with.

Does this make sense? If so, is it already possible to do this within SilverFast? If it does make sense and SilverFast doesn't permit it, can we expect a feature like this soon?

Thanks,
Carson