Texture Synthesis Given a texture, create more CS176: Texture - - PDF document

texture synthesis
SMART_READER_LITE
LIVE PREVIEW

Texture Synthesis Given a texture, create more CS176: Texture - - PDF document

Texture Synthesis Given a texture, create more CS176: Texture Synthesis All examples from Wei & Levoy CS 176 Winter 2011 CS 176 Winter 2011 1 2 Texture Synthesis Texture Synthesis Dont expect too much And amazing successes All


slide-1
SLIDE 1

CS176: Texture Synthesis

CS 176 Winter 2011

1

Texture Synthesis

Given a texture, create more

CS 176 Winter 2011

2

All examples from Wei & Levoy

Texture Synthesis

Don’t expect too much

CS 176 Winter 2011

3

All examples from Wei & Levoy

Texture Synthesis

And amazing successes

CS 176 Winter 2011

4

All examples from Wei & Levoy All examples from Ashikhmin

How does it work?

Big idea

 statistical assumptions:

 Markov random field model

t ti it d di it

CS 176 Winter 2011

5

 stationarity and ergodicity

 find pixels with similar neighbors

 scanline order (causal neighborhood)

 implementation: exhaustive search

In A Nutshell

Find most similar neighborhood

1 2 3 4 5 6 7 8 9 10

CS 176 Winter 2011

6

 high dimensional point/vector

11 12 1 2 3 4 5 6 7 8 9 10 11 12

slide-2
SLIDE 2

Details

Choices

 neighborhood size  hierarchy

d

CS 176 Winter 2011

7

 distance measure  acceleration structures  cut search space

Neighborhood Size

33 55 77

423 s 528 s 739 s

CS 176 Winter 2011

8

99 1111 4141

1020 s 1445 s 24350 s

Hierarchical Approach

Smaller neighborhoods ok

CS 176 Winter 2011

9

single res 55 3 levels 55 single res 1111

Applications

What are instances of synthesis?

 repair (inpainting)  image editing

l

CS 176 Winter 2011

10

 extrapolation  user control (introducing bias)

Resources

Papers

 Wei & Levoy

 Fast Texture Synthesis using TSVQ

A hikh i

CS 176 Winter 2011

11

 Ashikhmin

 Synthesizing Natural Textures

 check their web pages

Enhancements

Additional ideas

 store coordinates not values

 better for upsampling

l jitt t

CS 176 Winter 2011

12

 upsample, jitter, correct  multiple passes

 parallel subpasses

 search only shifted pixel nghbd.

slide-3
SLIDE 3

Enhancements

 Tong et al.: k-way coherence search

 preprocess exemplar with nearest list

CS 176 Winter 2011

13

 Lefebvre+Hoppe: appearance space

 distance to features as addl. data  PCA on neighborhoods

Transformed Exemplar

CS 176 Winter 2011

14

Example for 75D to 3D

CS 176 Winter 2011

15

Results

CS 176 Winter 2011

16

Results

CS 176 Winter 2011

17

Results

CS 176 Winter 2011

18