36
Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

Embed Size (px)

Citation preview

Page 1: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

Efficient Editing of Aged Object Textures

By:

Olivier Clément

Jocelyn Benoit

Eric Paquette

Multimedia Lab

Page 2: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

2

Multimedia Lab

Introduction

Realistic image synthesis Virtual reality, video games, special effects,

etc. Aging (or weathering)

Many effects Many objects Time consuming

Page 3: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

3

Multimedia Lab

IntroductionRedesign iterations

Produces theappropriate texture

Visualizes theappearance of an object

Reviewsthe result

Page 4: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

4

Multimedia Lab

Outline

Objectives Previous Work Aging Editing Process

Segmentation Phase Elimination Phase Reproduction Phase

Results and Limitations Conclusion

Page 5: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

5

Multimedia Lab

Objectives

To build a system To edit aging effects on textures To increase realism To reduce the amount of work Adapted for artists

adequate control interactive no complex parameters

Page 6: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

6

Multimedia Lab

Outline

Objectives Previous Work Aging Editing Process

Segmentation Phase Elimination Phase Reproduction Phase

Results and Limitations Conclusion

Page 7: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

7

Multimedia Lab

Previous Work

Physically based methods[Dorsey and Hanharan 2000; Merillou et al. 2001; O’Brien et al. 2002; etc.]

Highly realistic results but lengthy calculations Non-intuitive physical parameters

Empirical methods[Chain et al. 2005; Gobron and Chiba 2001; Paquette et al. 2002; etc.]

More intuitive parameters Both approaches

Do not provide the control required by artists Target a single aging effect

Aging methods

Page 8: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

8

Multimedia Lab

Previous Work

Image based[Gu et al. 2006; Wang et al. 2006; etc.]

Capture the time-varying aspects of the material Similar to our approach

Focus of our approach Simple capture process Adequate control

Aging methods

Page 9: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

9

Multimedia Lab

Outline

Objectives Previous Work Aging Editing Process

Segmentation Phase Elimination Phase Reproduction Phase

Results and Limitations Conclusion

Page 10: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

10

Multimedia Lab

Aging Editing Process

Source image Image, photograph Containing aging

effects Target aging mask

Binary image Desired pattern

Reproduction image New aging effects

Process overview

Page 11: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

11

Multimedia Lab

Aging Editing Process

Segmentation phase Semi-automatic Aged regions

Elimination phase Automatic Aging removed

Reproduction phase Automatic New aging effects

Phase description

Red

esig

n it

erat

ion

s

Page 12: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

12

Multimedia Lab

Aging Editing ProcessImages summary

Page 13: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

13

Multimedia Lab

Outline

Objectives Previous Work Aging Editing Process

Segmentation Phase Elimination Phase Reproduction Phase

Results and Limitations Conclusion

14

Page 14: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

14

Multimedia Lab

Segmentation Phase

Identifies aged regions

Could be done with Segmentation tools Image editing software

Stroke-based technique Lischinski et al. [2006]

Worked efficiently for semi-automatic identification

Page 15: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

15

Multimedia Lab

Segmentation PhaseStroke-base technique - Video

Page 16: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

16

Multimedia Lab

Outline

Objectives Previous Work Aging Editing Process

Segmentation Phase Elimination Phase Reproduction Phase

Results and Limitations Conclusion

17

Page 17: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

17

Multimedia Lab

Elimination Phase

Constrained texture synthesis Match the non-aged neighbourhood

Search using ANN library Arya et al. [1998]

The algorithm

best match

newbest match

Elimination image Source image

copy thepixel color

Page 18: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

18

Multimedia Lab

Elimination Phase

The boundary pixels Non-aged pixels in

their neighbourhood Must be filled first

The aged region is filled iteratively

Hole-filling

Page 19: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

19

Multimedia Lab

Outline

20

Objectives Previous Work Aging Editing Process

Segmentation Phase Elimination Phase Reproduction Phase

Results and Limitations Conclusion

Page 20: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

20

Multimedia Lab

Reproduction Phase

Extension of the elimination algorithm

Consider the aged / non-aged context

The new term

Page 21: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

21

Multimedia Lab

Reproduction Phase

Does not synthesize the entire image

Only specified regions

Iterative construction from multiple source images

Aging effects transfer and combination

Page 22: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

22

Multimedia Lab

Outline

Objectives Previous Work Aging Editing Process

Segmentation Phase Elimination Phase Reproduction Phase

Results and Limitations Conclusion

Page 23: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

23

Multimedia Lab

Results

Source image Elimination image Reproduction imageSource aging mask Target aging mask

Page 24: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

24

Multimedia Lab

ResultsSource image Elimination image Reproduction image

Page 25: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

25

Multimedia Lab

ResultsSource image Elimination image Reproduction image

Page 26: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

26

Multimedia Lab

ResultsSource image Aging masks Reproduction image

More results in thepaper and the video…

Page 27: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

27

Multimedia Lab

Results

User interaction is minimal Interactive computation time Efficient for redesign iterations

Efficiency

2.5 minutes - once25 seconds - once2 minutes

every iteration3 seconds

every iteration

Obtained on a PC with 3.2 GHz CPU and 3GB of RAM

Page 28: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

28

Multimedia Lab

Limitations

Apply only on surfaces No fractures or deformations

Camera-based texture acquisition Specular lighting Surface distortion

Current implementation Interactive on textures up to 512 x 512

Page 29: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

29

Multimedia Lab

Outline

Objectives Previous Work Aging Editing Process

Segmentation Phase Elimination Phase Reproduction Phase

Results and Limitations Conclusion

Page 30: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

30

Multimedia Lab

Conclusion

A framework To edit aging effects on textures To reduce the amount of work needed during

the redesign iterations Benefits

Appropriate for artists adequate control and interactivity no complex parameters

Works well for several types of aging effects

Page 31: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

31

Multimedia Lab

Conclusion

Synthesize the target aging mask For numerous regions Ex: scratches

Handle layers in effects combination Multiple effects over the same regions Ex: dirt on top of rust

Faster synthesis To handle higher resolution textures

Future work

Page 32: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

32

Multimedia Lab

? We would like to thank :

And all our reviewers…

Questions

Page 33: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

33

Multimedia Lab

Previous Work

Texture synthesis[Efros 1999; Hertzmann 2001; Kwatra 2003; Lefebvre 2006; Liang 2001; etc.]

Synthesis based on neighbourhood matching

Our system Extends from these algorithms Specializes for the aging context

Texture synthesis

Page 34: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

34

Multimedia Lab

Previous Work

Image analogies, Hertzmaan et al. [2001] The output image is completely synthesized Our approach uses a similar algorithm that

synthesize only regions of the output Our approach should be considered as an

extension

Texture synthesis

Page 35: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

35

Multimedia Lab

Elimination Phase

The replacement pixel is : Selected from the non-aged pixels of the source

image One of the best neighbourhood matches

The system seeks a replacement pixel that minimizes the following L2 norm :

The replacement pixel

Page 36: Efficient Editing of Aged Object Textures By: Olivier Clément Jocelyn Benoit Eric Paquette Multimedia Lab

36

Multimedia Lab

Elimination Phase

An exhaustive search would require processing time far from interactive

Thus, an approximation of the best match is found with the ANN library (Arya et al. [1998]) Approximate nearest neighbour searching algorithm

based on a kd-tree structure Our feature vector is composed of the RGB

components of the non-aged pixels around the pixel to replace

Interactivity