Anyone have an explanation of how the refraction component works?

Uberzev, I suppose you mean the refraction component, not filter, so I'm correcting the title.

The Refraction component simulates the following:

Imagine that you have a "bead" of transparent material (i.e. glass) with flat bottom and a non-flat (specified by Height input) top. "Stick" the Source input image to the flat bottom, specify the 1/IOR (index of refraction aka refraction coefficient indirectly mapped with scaling by 4.0, see your physics book for the explanation, or visit http://scienceworld.wolfram.com/physi...tion.html) with the Refraction input and look at the "bead" from the top -- that's the result.
Unfortunately, there's no ability to specify the IOR relations larger than one, so it's always air-to-surface transfer, not vice versa which may be helpful in simulating layered materials with help of surface filters.

I am under the impression that for the refraction component the "angle" between two (or four?) neighbouring pixel is measured - thus it behave different than the offset component which only looks at the individual pixel. But maybe I'm also totally wrong on this.

The normal map used for refraction calculations is approximated using 3-tap sampling (anisotropic, right/bottom-oriented, but it does not matter since our supersampling is isotropic and symmetric with centered sampling) using step which is equal to 0.001 pixels, respecting Size slider value. This dependence on Size creates unwanted effects when Image or Selection are used as inputs, but generally is very well suited for all procedural and bitmap-based components. The same algorithm is used for Surface normal map generation which is used then in lighting calculations.

Although this is a very old thread, is good to know better how the refraction works instead of only the explanation given in the FF Help

Quote
Sphinx. wrote: Does this mean that we should divide conventional refraction index values by 4 (in relation to what onyXMaster states above) to get the corresponding refraction?

I'm afraid refraction values used in FF have no direct relation to conventional refraction indices. I've asked the developers to look into the component code, and it seems the [0...100] slider is mapped to the IOR range of [infinity... 1].

I wonder what an IOR of infinity would mean in reality

I guess we could try set up a test scene in a 3D renderer to find out what values would match the IOR of water/glass ect. Like render a half of a glass sphere with a grid below it and tweak the refraction value in FF to match the distortion. Should work right?

Quote
GMM wrote: I have little understanding how it really works. The exact formula that is fed into the renderer is as follows: asin((1 - slider_value)*normal_projection_length).

Thanks, that makes sense - the slider is inverted like I expected. I'll see if I can figure out the relation to the conventional numbers

Quote
emme wrote: I wonder what an IOR of infinity would mean in reality I guess we could try set up a test scene in a 3D renderer to find out what values would match the IOR of water/glass ect. Like render a half of a glass sphere with a grid below it and tweak the refraction value in FF to match the distortion. Should work right?

IOR of infinity? A piece of black hole core would be involved here, I guess

Your experiment would definitely be a help, also to verify a solution a some point

Ok, here is a test. I rendered the reference images in Blender, then matched the refraction in FF. There is something wonky with the normal calculation at the top of the hemisphere in FF. Or it could be a problem with free ellipse / circular arc math. Other than that, the refractions match very closely with these values.

IOR 1.333 (water) -> 4.5
IOR 1.450 (glass) -> 5.5