TY - JOUR

T1 - Using Stochastic Modeling for Texture Generation

AU - Haruyama, Shinichiro

AU - Barsky, Brian A.

PY - 1984/3

Y1 - 1984/3

N2 - Blinn2produced wrinkled and bumpy textures by perturbing the direction of the surface normal vector, but his method required specific texture pattern data for perturbation. Noting that rough textures have inherently random structures, we have developed a new computer graphics method that uses stochastic modeling to generate highly realistic random textures. Stochastic modeling has been used in computer graphics by Fournier, Fussell, and Car-penter3to generate stochastic curves and surfaces. We use a similar technique except that we apply the stochastic function to the normal vectors instead of the surface position. We also extend the fractional Brownian motion (fBm) form of stochastic modeling by allowing different values of the self-similarity parameter h on different recursion levels. The self-similarity parameter determines how fast a stochastic factor decreases as the recursion level becomes deeper. If h is zero, then the fBm parameter is the same for all recursion levels. A large h creates a very smooth texture and a small h makes a very rough texture; this means that h can control the spatial frequency distribution. By using different self-similarity parameters for different recursion levels, we can adjust the roughness in more detail and even control the size of the wrinkles easily. Futhermore, we can control the height of the bumps in the texture by adjusting the standard deviation σ.

AB - Blinn2produced wrinkled and bumpy textures by perturbing the direction of the surface normal vector, but his method required specific texture pattern data for perturbation. Noting that rough textures have inherently random structures, we have developed a new computer graphics method that uses stochastic modeling to generate highly realistic random textures. Stochastic modeling has been used in computer graphics by Fournier, Fussell, and Car-penter3to generate stochastic curves and surfaces. We use a similar technique except that we apply the stochastic function to the normal vectors instead of the surface position. We also extend the fractional Brownian motion (fBm) form of stochastic modeling by allowing different values of the self-similarity parameter h on different recursion levels. The self-similarity parameter determines how fast a stochastic factor decreases as the recursion level becomes deeper. If h is zero, then the fBm parameter is the same for all recursion levels. A large h creates a very smooth texture and a small h makes a very rough texture; this means that h can control the spatial frequency distribution. By using different self-similarity parameters for different recursion levels, we can adjust the roughness in more detail and even control the size of the wrinkles easily. Futhermore, we can control the height of the bumps in the texture by adjusting the standard deviation σ.

UR - http://www.scopus.com/inward/record.url?scp=0021386180&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0021386180&partnerID=8YFLogxK

U2 - 10.1109/MCG.1984.276056

DO - 10.1109/MCG.1984.276056

M3 - Article

AN - SCOPUS:0021386180

VL - 4

SP - 7

EP - 19

JO - IEEE Computer Graphics and Applications

JF - IEEE Computer Graphics and Applications

SN - 0272-1716

IS - 3

ER -