Memory bandwidth is often the bottleneck for graphics systems, as the yearly performance growth rate of computing capability is much larger than that of bandwidth and latency for DRAM. Reading from texture memory is often the main consumer of bandwidth. Thus texture compression is widely used in graphics systems. S3TC (S3 Texture Compression) is the most common texture compression scheme, including compression methods for the alpha channel and color channel of texture images. Although more than 10 years have passed since Iourcha et al. presented the S3TC compression algorithms, it seems that no other texture compression scheme could challenge the supremacy of S3TC.
Under the guidance of Professor Weiwu Hu, PhD student Yifei Jiang presented a novel alpha compression scheme, called EAC (Enhanced Alpha Compression), as an improvement to that of S3TC. EAC defines new compression data formats, and employs clustering algorithms combined with linear interpolation method used in S3TC to compress 16 input alpha values to 64 bits. The compressed block has three possible formats, corresponding to two process modes in decompression. Results show that EAC improves the average PSNR with about 0.2 dB compared to alpha compression of S3TC/DXT5 over a set of test images. Moreover, the average cell area of hardware implementation is reduced by about 25%, and the power consumption is lowered by 21.66%-26.75%.
The work entitled “Alpha Compression with Variable Data Formats” was reported at the 32nd annual conference of the European Association for Computer Graphics (Eurographics 2011), held in the town of Llandudno in North Wales, UK from 10th to 15th April, 2011.
“I very much enjoyed reading the paper. I like the idea of improving the compression of the alpha channel this way. The paper raises a topic that requires re-thinking (of current texture compression) and (current method) really needs improvements. Their improvements should be useful in practice.” “The paper is an interesting research that could lead to better texture compression algorithms in the future.” The work earned high remarks form the Eurographics 2011 reviewers, and was believed to give an inspiration to new texture compression data formats, from which innovative texture compression standard could derive. The paper will be delivered by Eurographics digital library at www.diglib.eg.org.
AUTHOR CONTACT:
Yifei Jiang
Institute of Computing Technology, Chinese Academy of Sciences
Beijing, China 100190.
Phone: 86-10-6261313; E-mail: e_fly@ict.ac.cn
downloadFile