Game Development Reference
In-Depth Information
Figure 4.4. Close-ups demonstrating the quality of our edge-directed reconstruction
filter in low-dynamic-range sRGB input. As expected, the largest amount of error
occurs on edges and areas with noisy content, but it is still low enough to not be visible
to the human visual system. The error is measured as the absolute difference in the
RGB colors between the original and the compressed data.
The quality of the framebuffer is very important, therefore the reconstruction
should be robust enough to handle the most challenging cases, like high-frequency
content and strong chrominance transitions, without introducing any visible arti-
facts. Figure 4.4 demonstrates that these challenging cases are handled without
any visual artifacts by the edge-directed filter. High-dynamic-range (HDR) con-
tent provides some extra challenges to chroma subsampling schemes, since the
HDR of the luminance tends to exaggerate any “chrominance leaking” on edges
with high dynamic contrast. The edge-directed nature of the reconstruction in
our method prevents the appearance of any chrominance leaks, even on edges
with extremely high dynamic contrast, as the one shown in Figure 4.5. For the
test in this figure we have integrated our technique to a well-known demo by Emil
Persson, combining our compact format with multisample antialiasing (MSAA),
HDR render targets, and proper tone mapping before the MSAA resolve.
To test the temporal stability of our method with moving scenes, we have
conducted two experiments. In the first experiment we have used the demo of
Figure 4.5 and moved the camera around. But since the content of a real game
might be more demanding, we have also encoded a sequence of framebuffers from
a real game with our method. In both tests our method appears to be temporally
Search Nedrilad ::

Custom Search