1. Computer Generated Watercolor
Curtis, Anderson, Seims, Fleisher, Salesin
SIGGRAPH 1997
Presented by
Yann SEMET
Universite of Illinois at Urbana Champaign
Universite de Technologie de Compiegne
2. Background
NPR
Purpose : aesthetic rather than
technical
Artificial art ?
23. Fluid simulation II
Parameters of the simulation :
Wet-area mask : M
Velocities : u,v
Pressure : p
Concentration : gk
Height of paper : h
Physical properties : density, staining power,
granularity, etc.
Fluid properties : saturation, capacity, etc.
24. Paper simulation
Supposedly : shape of every fiber
matters
A simpler model : a height field
Generation : Perlin’s noise and Worley’s
cellular textures
25. Main loop
For each time step
Move Water
Update velocities
Relax Divergence
Flow Outward
Move Pigment
Transfer Pigment
Simulate Capillary Flow
26. Conditions for realism
Flow must be constrained so water
remains within M
Surplus of water causes flow outward
Flow must be damped to minimize
oscillating waves
Flow is perturbed by texture of paper
Local changes have global effects
Outward flow to darken edges
27. Rendering : Kubelka-Munk
For each pigment, 2 coeff. Per RGB layer :
K : absorbtion
S : scattering
Supposedly : K and S are measured
Here : user provides Rw and Rb
28. Types of paints
Opaque (e.g. Indian Red)
Transparent (e.g. Quinacridone Rose)
Interference (e.g. Interference Lilac)
Different hues (e.g. Hansa Yellow)
29. Optical compositing
Compute R and T :
Then compose :
Weight relatively to relative thicknesses
30. Discussion of the KM model
Assumptions partially satisfied :
Identical refractive indices
Random orientation of pigments
Diffuse illumination
1 wavelength at a time
No chemical interaction
Works surprisingly well !
OK, because we’re looking for appearance,
not actual modeling