So I am having some fun trying to fake some SSAO using only the depth buffer (as described in the
Journal of Ysaneya). As is to be expected, my results so far are somewhat less than I would have desired
Here is a (very preliminary) screenshot of just the AO term that is to be multiplied with the scene - At the moment it looks more like an edge-blur than an AO term, but with time:
larger image
Currently I take the depth buffer, take the absolute difference of a pixel's depth and its neighbours, and use that factor to set the radius of 12-sample Poisson disk blur, multiplying the sample distance by the radius, and running 2 passes. This is giving fairly decent results, with out a huge performance hit, but I think a better blur might be possible, so if anyone has suggestions on that, they would be most welcome...
Also, I really need a more complex scene to test with this - but I am no artist
If anyone has an indoor scene I could borrow for these tests, I would be overjoyed!