Linear elevation vs gaussian blur

What if you turn off the Gaussian setting?

Otherwise you might want to post a snippet.

i dont know why, but when i go to save your snippet, zoltan, it changes the .ffxml to just .xml in the save box before i save it. other filters posted on the forum maintain the .ffxml (i just tried it) so, i dont know what's going on with yours. might want to look at it.

I figured it out. Please stand by.

ok, nevermind

I ran into a bug

In the mean time if you want a soft disk try a round frame component with a circle curve plugged in.

By using a Gaussian curve I was able to turn a soft disc into something like a blurred disc looks like. Not extremely precise, but close enough.
It looks like this.

But now I have no idea how to reverse the process. (Ie: making a soft disc from the blurred one.) Inverting or flipping the curve didn't help.

Here's the EXACT curve you need:



I've done my best to recreate it...

unGaussian.ffxml

From a gamma and a gaussian curve's difference I managed to create a curve which made the blurred disc kinda linear. (The top has the greatest error, it isn't pointy.)

I thought I nailed it until I tried on a blurred vertical bar: it doesn't really work with that. It's not a steady curve but bumpy. It is probably caused by the fact that each pixel column has the same value. The gaussian blur will find the same value vertically regardless of the radius. Or something.

So what worked for the disc, looks crap on a bar. (Not to mention that I'd need to adjust 5 parameters whenever the blur radius changes.)

So I started to wonder if it is a solvable problem. Or maybe since the gaussian blur samples from all around, it produces many different elevations, so there is no "one curve fits all" way to solve this.

Here is where I'm at right now

DeGauss

What you see with the default settings is the grossly exaggerated difference between the reference soft disc and a blurred one with a tone curve applied. The top is still too soft, but I could live with that.

However, if you uncheck Disc, so the input image becomes a bar, the difference skyrockets.

One problem with this test method is that I rely on the soft disc to check the elevation against. This means that when changing the blur radius, the white disc's radius also need to be changed, so it covers the same area as the reference soft disc.
I'm working on a way to visualize the cross section of a chosen pixel row in the result image. This way I could see clearly how parameters change the elevation, without the need of a reference (which does not exist when using an arbitrary shape).

The controls are prioritized, so the top 4 is what I massaged most of the time.

Ps: Does this file lose its .ffxml extension as well? It works for me, although instead of opening it in FF, FFox just shows the plain text. o_O

Weird. Maybe some encoding issue...

Anyway, here is the latest:

DeGauss2

(I used bits from Mike Blackney's pixel color and uberzev's Curve to shape filters.)

SourceType picks from Disc/Bar/Noise1/Noise2
NoCorrection removes the "degauss" curve
CrossSection shows the elevation at a given pixel row
CrossSectionPos changes the sampled row

When the base shape is not a simple one then it gets kinda confusing. In case of a noise I don't really know how a linear elevation should look like.

This is how it looks.