physically correct subsurface scattering basics

Fd80f81596aa1cf809ceb1c2077e190b
0
rouncer 104 Aug 22, 2012 at 14:25

so say i was going to photon blast my way the light side of the model all the way to the visible side, how does the light get there collision to collision. i figure ive got to understand this at least or i wont have a chance at it ever at all.

I realize it would be similar to diffuse reflection, like a collision of a ray would produce many rays - and maybe some reversed randomized path tracing method from visible side to the light might be a better implementtion - which would give a correct result after averaging some high enough sample count.

I dont even know how to start, I realize it definitely has a vectoral property, as it has before its even glanced off the surface to begin with. First thing I worked out that draw me to the conclusion I know nothing, was what do I even do to begin with? Just taking the diffuse reflectance and starting with that as a factor sounds completely wrong, Im guessing colliding to begin with and colliding ever after could be the exact same method, its got to be some dependancy of how far apart you make your collisions, like youd think absorption and angle change would have to do with how far you make your collisions apart as an optimization.

Id like it detailed enough a method, so you could put non scattering substances inside, or even take into account an uneven medium.

Im thinking possibly it could be simple but there may be some strange words involved.

Anyway, if anyone could help id be very happy… im tired of having basic gi and no scattering to go with it - my lux 3d renderer is never going to happen at this rate.

5 Replies

Please log in or register to post a reply.

Fd80f81596aa1cf809ceb1c2077e190b
0
rouncer 104 Aug 22, 2012 at 15:22

I just had a maybe cool thought. is it like just passing through and eventually all of it reflects off a collection of randomly directioned minute faces due to mixture thickness?

Fd80f81596aa1cf809ceb1c2077e190b
0
rouncer 104 Aug 22, 2012 at 16:34

another cool thing, is actually its a bit like completely chaotic global illumination path tracing, add light when you hit a lit portion for so many chaos samples. sorta like a form of micro path tracing.

A8433b04cb41dd57113740b779f61acb
0
Reedbeta 167 Aug 22, 2012 at 16:40

You’ve got the right intuition; it does have to do with how far apart the collisions occur. Technically this is called the mean free path. The shorter the mean free path, the more the light gets scattered and the denser the medium appears. If you read anything about volume scattering they’ll talk about “scattering coefficients” - that’s basically 1.0 / the mean free path. Red, green, and blue (or really all wavelengths) could all have a different mean free path.

So the light goes through and gets randomly scattered with a certain probability per unit length. Each time it’s scattered it ends up going in a random other direction. Just as there’s a BRDF for a surface that describes the probability of getting scattered in different directions, likewise there’s a thing called the phase function that describes the probability of scattering in different directions for a volume. The Henyey-Greenstein phase function is a common one. For subsurface I’d imagine you could just use isotropic scattering (all goes in directions equally) as a good approximation though.

Fd80f81596aa1cf809ceb1c2077e190b
0
rouncer 104 Aug 22, 2012 at 17:47

thanks man, that extra info is really good… this is looking a whole lot less mystified to me now… whenever something looks complex, you know its kindof a lie, all that maths i dont know was always getting in the way of me understanding the papers. i just had to finally use my Socratic mind and realize it really is a landslide of simple, everything is.

cheers man *clink*

6eaf0e08fe36b2c23ca096562dd7a8b7
0
__________Smile_ 101 Aug 23, 2012 at 01:19

I think you can use this image for reference (right part shows exact ray path in media). Each turn point is something like Rayleigh scattering (you have to account polarization to do in right).