Fluid Simulation as a Tool for Painterly Animation

1
Fluid Simulation as a Tool for Painterly Animation
Other details   Our stroke based rendering system
follows the basic template introduced in
Processing images and video for an impressionist
effect. We use a quality controlled Delaunay
triangulation to control stroke density. We clip
strokes to region boundaries, and apply random
permutations to stroke color and width. We can
create videos by processing each frame of a
source video clip, or by generating multiple
renderings from a single source image.   We draw
curved brush strokes by advecting a point through
the brush alignment field.   We use vorticity
confinement forces to exaggerate the turbulent
effects in our fluid fields.   The brush stroke
settings can vary between regions we may use
different stroke textures and permutation ranges
in different regions.
The idea We love the turbulent brush stroke
patterns in Starry Night, and we wanted to
develop NPR algorithms that would give us a
similar effect. We combined a stroke based
rendering system with vector fields from a
semi-Lagrangian fluid simulation.
Image Regions and Vector Fields   We split the
source image into regions using a color space
clustering method. For each region, we specify a
vector field which will control stroke alignment.
If were making videos, we define a second field
that controls stroke motion. The types of fields
that we use are   TPS Fields We use the region
definitions to imply a set of vectors tangent to
the region boundaries. Those vectors become
control points for a pair of thin plate splines.
The result is a smooth vector field that is
neatly aligned to the region boundaries.   Fluid
Velocity Fields We use a semi-Lagrangian fluid
simulation to create a dynamic velocity field.
Currently were using forces derived from the TPS
control points to initialize the fields, but
were looking for other interesting ways to setup
or influence the fluid fields.   Hybrid Fields
Fields defined as a weighted sum of the fluid and
TPS fields, where the weights are a function of
the distance from the nearest TPS control
point.   Optical Flow Fields When working with
video, advecting the strokes via an optical flow
field can make them seem to stick to objects in
the video.
Fluid velocity fields for the poppy demo
TPS Fields for the poppy demo
Application binaries, source code, and additional
demos and discussion available at
http//www.sccs.swarthmore.edu/users/03/sven/npr
? Fluid fields have interesting eddies and evolve
smoothly over time. ? TPS fields are aligned to
region boundaries. ? With hybrid fluid/TPS
fields we try to get the best of both worlds.
Sven C. Olsen (sven2718_at_verizon.net)Bruce A.
Maxwell (maxwell_at_swarthmore.edu) Swarthmore
College