A compact and editable representation for measured BRDFs

University of Bonn, Technical Report number CG-2010-1, Dec. 2010
 

Abstract

Data-driven Bidirectional Reflectance Distribution Functions (BRDFs) are capable of representing a very large space of materials faithfully. However, these approaches commonly suffer from two major problems. First, the measured data is often very large and thus compression techniques are necessary to considerably decrease the data size with an acceptable degradation in perceptual quality. Second, while editing measured datasets is often difficult, measuring all required datasets might easily become impractical. In this paper, we show that a perceptually very accurate representation for isotropic BRDFs can be obtained by minimizing a relative squared error during a PARAFAC decomposition of the BRDF tensor. This representation allows for efficient real time rendering of a large number of materials at a time. Furthermore, we present an editing technique which performs perceptually plausible interpolations between BRDFs by using dynamic time warping to align BRDF features during the interpolation. Using our compression technique and precomputed warping information, the interpolation between several materials can be performed in realtime on a GPU. This makes a very intuitive interface possible which allows for interactive exploration of the space spanned by a database of BRDFs.

Images

Download Paper

Download Paper

Bibtex

@TECHREPORT{ruiters-2010-compact-BRDFs,
       author = {Ruiters, Roland and Klein, Reinhard},
        title = {A compact and editable representation for measured BRDFs},
       number = {CG-2010-1},
         year = {2010},
        month = dec,
  institution = {University of Bonn},
     abstract = {Data-driven Bidirectional Reflectance Distribution Functions (BRDFs) are capable of representing a
                 very large space of materials faithfully. However, these approaches commonly suffer from two major
                 problems. First, the measured data is often very large and thus compression techniques are necessary
                 to considerably decrease the data size with an acceptable degradation in perceptual quality. Second,
                 while editing measured datasets is often difficult, measuring all required datasets might easily
                 become impractical. In this paper, we show that a perceptually very accurate representation for
                 isotropic BRDFs can be obtained by minimizing a relative squared error during a PARAFAC
                 decomposition of the BRDF tensor. This representation allows for efficient real time rendering of a
                 large number of materials at a time. Furthermore, we present an editing technique which performs
                 perceptually plausible interpolations between BRDFs by using dynamic time warping to align BRDF
                 features during the interpolation. Using our compression technique and precomputed warping
                 information, the interpolation between several materials can be performed in realtime on a GPU. This
                 makes a very intuitive interface possible which allows for interactive exploration of the space
                 spanned by a database of BRDFs.}
}