Compression of Synthesized Textures

In spite of graphics hardware advancements, graphics memory is still a scarce resource for usual applications. Besides, for most raster-based applications, the available bandwidth is one important limiting factor for increasing performance in the system. Texture compression addresses both of these problems. We introduce a new technique for compression of textures synthesized from samples. The compressed texture stores the sample plus encoded data, gathered during synthesis, which enables real-time decompression of large textures. With this scheme we are able to achieve high compression rates. Our solution explores a spectrum of textures where general texture compression schemes achieve less than optimal compression rates. These are usually textures with repeating patterns, regular or near-regular ones, and stochastic ones, the exactly types of textures where texture synthesis algorithms perform well. We also present analytical formulae for our compression scheme that allows an exact computation of compression rates achieved.