PDA

View Full Version : Meaning of "Soft" "Sharp" in Pixel Filters ?



grabiller
07-31-2014, 03:39 AM
Hello,

It would be interesting to have more technical details about the meaning of "Soft"/"Sharp" pixel filters.

For instance, 'standard' kernel filter width in computer graphics for Mitchell is 4x4.

Does it means Mitchell "Soft" has a filter width of 3x3 and Mitchell "Soft" of 5x5 or something ?

And does it means that "Mitchell" (without Soft/Sharp) is indeed 4x4 ?

Or is the Sharp/Soft factor is acting on the actual sharpness parameter of an adjustable Mitchell filter. This is important to know as, for instance, instead of asking for a Blackman-Harris 3x3, we could use Mitchell 4x4 with 0.4 sharpness which is quite equivalent.

Cheers,
Guy.

3DGFXStudios
08-01-2014, 02:54 AM
From what I've heard are pixel filters in every program a little different and not comparable with each other. At least they will not give you the exact same result. Every program has it's own interpretation of let's say a Mitchell filter.

grabiller
08-01-2014, 06:18 AM
From what I've heard are pixel filters in every program a little different and not comparable with each other. At least they will not give you the exact same result. Every program has it's own interpretation of let's say a Mitchell filter.

Mathematically, a Mitchell filter is a Mitchell filter, there is only one way to compute it. However, there are several "factors" than can be customized by the developers, especially the kernel width, which is often exposed (but not in Lightwave) and a sharpness factor.

When you say every program has its own interpretation I would rather say "use its own factors values". Some softwares expose those (or some of those) factors, some others do not.

The thing is the users would like to know which values are used for those factors (and perhaps change it).

Hiding values of these factors is like saying to the user: your are not smart enough, you don't need to know.

3DGFXStudios
08-01-2014, 12:18 PM
I agree with you on that.
I think there was a thread here on the forum that discusses this issue not so long ago.