In 1999, Tumblin and Turk published a paper on using a low curvature image simplifier, or LCIS filter, to do HDR compression. For my project, I extended the LCIS filter to work in 3D and applied it to an HDR animation. The expectation was that this filter would allow for smooth HDR compression across the animation. Unfortunately, the results were less than pleasing.
When one performs HDR animation using the 2D LCIS filter on each frame of an HDR animation in isolation, a number of inconsistancies appear. In particular, "spots" appear and disappear near specularities. I initially believed that these "spots" would be smoothed from the animation when the HDR compression was performed with a 3D LCIS filter. After implementing both the 2D and 3D filters and running tests, I found the results to be almost identical (see the two videos posted below).
The similarities between results of HDR compression when using the 2D and 3D LCIS filters are not surprising once the filters are understood. The purpose of the LCIS filter is to smooth an image while preserving edges. Yet, Tumblin and Turk perform HDR compression by preserving, and effectively magnifying, the "detail" that is removed in the smoothing process. Thus, using a 3D LCIS filter preserves, and once again magnifies, the differences from one frame to the next in an HDR animation.
HDR Animation Compression using a 2D LCIS Filter