Self-Similar Texture for Coherent Line Stylization, NPAR2010

Pierre Bénard ……………… Grenoble University

Forrester Cole ……………… MIT CSAIL

Aleksey Golovinskiy …… Princeton University

Adam Finkelstein

Self-Similar Texture

for Coherent Line Stylization

Line stylization

2

Pablo Picasso, 1949Lithograph for the International Peace Congress

Andy Wakeley - Cheetah bounding sketch

IFÖ Cera wall mounted bidet 4016

Line stylization

3

0

l

Brush path

Line texture

Contributions

4

Coherent Parameterizationof View-Dependant Lines

Self-Similar Line Artmaps

Mapping Policies

5

• Stretching

• Tiling

Mapping Policies

6

• Stretching

Temporally coherent

Stretch or compress the texture

• Tiling

Mapping Policies

7

Mapping Policies

8

• Stretching

Temporally coherent

Stretch or compress the texture

• Tiling

Preserve the texture characteristics

Incoherent sliding

Mapping Policies

9

• Artmaps, Klein et al. 2000

• Tonal art map, Praun et al. 2001

Coherent transition

Blending artifacts

Restricted zoom

Mapping Policies:

Previous work

10

• Texture Fractalization, Cunzi et al. 2003 / Bénard et al. 2009

Infinite zoom

Pattern degradation

Mapping Policies: Previous work

11

+

Mapping Policies

12

• Stretching

Temporally coherent

Stretch or compress the texture

• Tiling

Preserve the texture characteristics

Incoherent sliding

• Fading / Artmaps

Smooth and coherent transitions

Blending artifacts

Mapping Policies

13

SLAMs

14

SLAM Definition

15

Lmax / 2

Lmax

16

SLAM Properties1) Density

2) Smooth variation

17

SLAM Properties3) Self-Similarity

18

SLAM Properties3) Self-Similarity

SLAM Properties

19

l=0.5

l=0

l=1

Lmax

Artmap Construction

20

• Manual:“Styles” in Google SketchUp (up to 7 levels)

• Multi-scale Texture synthesis, Han et al. 2008

Infinite zoom

Coherence

Pattern at multiple scales

Costly synthesison the fly

Previous work

21

SLAM Synthesis by Example

22

• Parametric texture synthesis, Portilla and Simoncelli 2000

• Property: (empirical observation, no formal proof)

small change in seed small change in output

SLAM Synthesis by Example

23

Input: white noise seed

Seed Pyramid

…… Input: texture example

Output: SLAM

……

Parametric Texture Synthesis

Seed Pyramid Construction

24

=

Wsource Wf

Seed Pyramid

Wf

[WfWf]

0.5

1

α Wf + (1-α) [WfWf]

α=2 -1

Lmax

Lmax

Lmax / 2

SLAM Synthesis by Example

25

Results: synthesis

26

Examples Synthesized textures

Results: live demo

27

Contributions

28

Self-Similar Line Artmaps

Coherent Parameterizationof View-Dependant Lines

3D Lines

29

• Creases• Ridges and valleys

• Silhouettes• Suggestive contours• Apparent ridges

Fixed Lines View-Dependant Lines

• Coherent Stylized Silhouettes,Kalnins et al. 2003

Very good coherence

Tradeoff between 2D and 3D

Complex and brittle

Simpler optimization for 2D arc-length parameterization

Previous work

30

• Parameterization: T(s) = s + = slope = phases = 2D arc-length

• Goal:

find ( , ) such that T(s) best fit the parameterization of the previous frame.

• 3 steps process in 2D screen-space

Screen-space coherence

31

1. Propagate Parameters

3. Render Strokes

Pipeline

32

2. Update Parameters

frame f

frame f+1

Parameters propagation

34

si

p

d

• Sample visible paths: si

• Splat parameterization:ti = si +

• Weighted by:wi(p)= clamp(1-d/,0,1)

exact parameterization along the spine of the line

• Contrary to Kalnins et al. 2003:

No item buffer (aliasing artifacts)

No local search

Additional controls:

– Size of the sprite: propagation vs. overlapping

– Weights: partial visibility / z-depth

Parameters propagation

35

1. Propagate Parameters

3. Render Strokes

Pipeline

36

2. Update Parameters

frame f

frame f+1

Parameters update

38

• RANSAC fitting [Fischer and Bolles 1981]

• Similar in spirit to Bourdev, 1998Mix information from multiple paths

Popping when paths merge

Parameters update

39

1. Propagate Parameters

3. Render Strokes

Pipeline

40

2. Update Parameters

frame f

frame f+1

Results

41

Results

42

• Synthesized texture quality

• Parameterization propagation2D screen-space buffer insufficient

Limitations and future work

43

SLAMs provides:

• Temporally coherent line stylization

• For 2D and 3D animations

• Including view-dependant features with our screen-space coherent parameterization

Conclusions

44

SLAMs synthesis source code and dpix viewer:

http://www.cs.princeton.edu/gfx/proj/dpix

Thank you for your attention

45

46

Results: 2D animation

47

Results: fixed 3D lines

48