Texture Compression with Variable Data Formats

Abstract

Texture compression is an important technique widely adopted in graphics processing units (GPUs) nowadays to reduce memory bandwidth consumption. However, modern texture compression schemes cannot generate satisfactory visual qualities for both alpha channel and color channel of texture images. In this paper we present a novel texture compression scheme, called CATC (Complexity-Aware Texture Compression), based on the insight into the essential differences between transparency and color. CATC defines new texture compression data formats and compresses the alpha channel and color channel of texture image flexibly. CATC employs clustering algorithm combined with the linear interpolation method to compress the alpha channel of texture images. If the alpha values are of low complexity, CATC takes advantage of the free bits in compressed block of alpha channel to compress color channel of texture image. While keeping the same compression ratio as the de facto standard texture compression scheme, CATC improves both the two channels' compression quality. The average PSNR score of alpha channel is improved by about 0.2 dB, and the average PSNR score of color channel can be increased by up to 7.17 dB over a set of test images, which makes CATC a better substitute for the standard texture compression scheme.