Polynomial Wavelet Trees for Bidirectional Texture Functions

Polynomial Wavelet Tree is a new tool for accurate and efficient compression of BTFs. The key idea is to separate directional and spatial variations by projecting the spatial BTF domain (i.e., the light-dependent textures) onto a wavelet basis and to approximate these light-dependent wavelet coefficients with a polynomial function. Fitting wavelet coefficients instead of data themselves is a more efficient approximation compared to previous solutions since low frequency light transitions are smooth and higher frequency coefficients can be quantized with less importance. This wavelet projection and the light-dependent polynomial approximation is done for each view-point. Our solution is also designed for efficient high quality materials rendering on GPU.

@InProceedings\{BBGS08,
  author       = "Baril, Jerome and Boubekeur, Tamy and Gioia, Patrick and Schlick, Christophe",
  title        = "Polynomial Wavelet Trees for Bidirectional Texture Functions",
  booktitle    = "ACM Siggraph 2008 - Talk Program",
  month        = "aug",
  year         = "2008",
  organization = "ACM",
  url          = "http://iparla.inria.fr/publications/2008/BBGS08"
}