Game Development Reference
In-Depth Information
45
0
Figure 4.1. Using our method, color images can be rasterized directly using only two
framebuffer channels: original uncompressed framebuffer stored using three color chan-
nels (left), and compressed framebuffer using two color channels in the YC o C g space
(right). The compressed frame appears indistinguishable from the original one. Inset:
Heat map visualizing the compression error (47.02 dB PSNR).
In this chapter we describe a practical lossy framebuffer compression scheme,
based on chrominance subsampling, suitable for existing commodity GPUs and
APIs. Using our method, a color image can be rasterized using only two frame-
buffer channels, instead of three, thus reducing the storage and, more importantly,
the bandwidth requirements of the rasterization process, a fundamental operation
in computer graphics. This reduction in memory footprint can be valuable when
implementing various rendering pipelines. Our method is compatible with both
forward and deferred rendering, it does not affect the effectiveness of any lossless
compression by the hardware, and it can be used with other lossy schemes, like
the recently proposed surface-based antialiasing (SBAA) [Salvi and Vidimce 12],
to further decrease the total storage and bandwidth consumption.
While our method is lossy, it does not result in any visible quality degradation
of the final rendered image, as shown in Figure 4.1. Furthermore, our measure-
ments in Section 4.7 indicate that in many cases it provides a rather impressive
improvement on the GPU fill rate.
4.2
Color Space Conversion
The human visual system is more sensitive to spatial variations of luminance in-
tensity than chrominance. This fact has been exploited by many image and video
coding systems, such as JPEG and MPEG, in order to improve compression rates,
 
 
Search Nedrilad ::




Custom Search