This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Anisotropic Noise Alexander Goldberg Matthias Zwicker Fredo Durand University of California, San Diego MIT CSAIL Anisotropic noise Isotropic filtering, no distortion compensation (a) (b) (a) (b) Figure 1: We present a technique for fast, high quality rendering of noise textures in interactive applications. We perform anisotropic filtering, which leads to higher image quality compared to isotropic filtering, as shown by the difference between close-ups (a). Our method uses 2D noise tiles and we present a technique to compensate for parametric distortions to achieve similar effects as with solid texturing. Our technique avoids texture distortions as shown by the difference between close-ups (b). Abstract Programmable graphics hardware makes it possible to generate pro- cedural noise textures on the fly for interactive rendering. However, filtering and antialiasing procedural noise involves a tradeoff be- tween aliasing artifacts and loss of detail. In this paper we present a technique, targeted at interactive applications, that provides high- quality anisotropic filtering for noise textures. We generate noise tiles directly in the frequency domain by partitioning the frequency domain into oriented subbands. We then compute weighted sums of the subband textures to accurately approximate noise with a de- sired spectrum. This allows us to achieve high-quality anisotropic filtering. Our approach is based solely on 2D textures, avoiding the memory overhead of techniques based on 3D noise tiles. We de- vise a technique to compensate for texture distortions to generate uniform noise on arbitrary meshes. We develop a GPU-based im- plementation of our technique that achieves similar rendering per- formance as state-of-the-art algorithms for procedural noise. In ad- dition, it provides anisotropic filtering and achieves superior image quality. 1 Introduction Noise functions are widely used in computer graphics to efficiently generate complex textures resembling natural phenomena. Essen- tially, these textures consist of a sum of band-limited noise images. By applying carefully designed functions to the noise bands, one can produce a variety of textures that resemble wood, marble, or clouds, etc. [Peachey 2003]. Noise functions are also often added to other textures to provide detail and a more natural look. Perlin  originally proposed to generate the noise bands pro- cedurally. His approach amounts to evaluating a low-pass filter on the fly when the texture is sampled. With programmable graph- ics hardware it is possible to use procedural noise for interactive rendering [Mine and Neyret 1999; Hart 2001; Green 2005; Olano 2005]. This is desirable because only minimal texture memory must be allocated for noise evaluation. However, it is difficult to perform high-quality antialiasing of procedural noise. Antialiasing is typ- ically achieved by truncating the noise frequencies to the Nyquist limit of the display. In theory, this can be done simply by omittinglimit of the display....
View Full Document
- Spring '08