Résumé: Texture mapping is an important operation in high-quality computer graphics applications. The principles of image "ltering are well established and understood and several high-quality texture "ltering algorithms have been developed. However, these have tended to be either o!-line software implementations or based on high-end computer graphics rendering systems. Current generation PC-based graphics acceleration utilises texture pre-"ltering techniques based upon an isotropic "lter kernel. The most common implementations are those based on MIP-map texture storage and bilinear or trilinear interpolation. Such "lters give a reduction in image aliasing at the expense of introducing blurring. As acceleration hardware performance improves, tolerance of such compromises is falling leading to the adoption of ` anisotropic a "ltering solutions. In this paper we present a detailed description of a low-cost implementation of an anisotropic "ltering unit for use within a contemporary graphics pipeline. The "lter is shown to produce improved visible results for only a modest increase in hardware cost.

Langue: Anglais