PORTRAIT PAINTING USING ACTIVE TEMPLATES Mingtian Zhao, Song-Chun - - PowerPoint PPT Presentation

portrait painting
SMART_READER_LITE
LIVE PREVIEW

PORTRAIT PAINTING USING ACTIVE TEMPLATES Mingtian Zhao, Song-Chun - - PowerPoint PPT Presentation

PORTRAIT PAINTING USING ACTIVE TEMPLATES Mingtian Zhao, Song-Chun Zhu University of California, Los Angeles Lotus Hill Institute http://bit.ly/portrait-painting Portrait Painting Particularity Structural Acute perception by human


slide-1
SLIDE 1

Mingtian Zhao, Song-Chun Zhu University of California, Los Angeles Lotus Hill Institute http://bit.ly/portrait-painting

PORTRAIT PAINTING USING ACTIVE TEMPLATES

slide-2
SLIDE 2

Portrait Painting

  • Particularity
  • Structural
  • Acute perception by human vision
  • Special Practices
  • Large and decisive brush strokes
  • Vibrant colors
  • Not so easy as painting generic objects
slide-3
SLIDE 3

Painterly Rendering of Human Portraits

  • Existing Methods
  • Strategic [Hertzmann 98; Hays and Essa 04]
  • Content-Oriented [Zeng et al. 09]
  • Stochastic Processes [Zhao and Zhu 11]
slide-4
SLIDE 4

Painterly Rendering of Human Portraits

  • Existing Methods
  • Strategic [Hertzmann 98; Hays and Essa 04]
  • Content-Oriented [Zeng et al. 09]
  • Stochastic Processes [Zhao and Zhu 11]
  • Challenges
  • Weak data vs. Strong prior
  • Large and decisive strokes: Structural details
  • Vibrant colors: Not present in original photographs
  • Our Approach
  • Example-Based Templates (Dictionary)
slide-5
SLIDE 5

Portrait Painting Dictionary

slide-6
SLIDE 6

Portrait Painting Dictionary

slide-7
SLIDE 7

Portrait Painting Dictionary

slide-8
SLIDE 8

Rendering Pipeline

Stroke Position Transformation Stroke Color Shift

slide-9
SLIDE 9

Rendering With Different Templates (Fig.1)

slide-10
SLIDE 10

Active Portrait Template

  • Brush Stroke Model
  • [Zeng et al. 09]
  • Quadrilateral Mesh
  • Thin-Plate Spline (TPS)
  • Facial Structure Model
  • Active Appearance Model (AAM)
  • [Cootes 01]
  • 83 Landmark Points
slide-11
SLIDE 11

Rendering

  • Template Selection
  • Compute AAM landmarks for input image
  • Search for best shape and color matching in the dictionary
  • Report 10 candidates for user selection
  • Shape distances in 5-D PCA space
  • Color distances in 25-D PCA space (Eigenface distance)
slide-12
SLIDE 12

Rendering

  • Stroke Position Transformation
  • TPS transformation defined by pairs of AAM landmarks
  • Stroke Color Shift
  • Objective: To preserve contrast among neighboring strokes
  • Inspired by color transfer of natural images [Reinhard et al. 2001]
  • In CIELCH color space
slide-13
SLIDE 13

Experiments (Fig.8)

slide-14
SLIDE 14

Experiments (Fig.10)

slide-15
SLIDE 15

Experiments (Fig.9)

slide-16
SLIDE 16

Experiments (Fig.11)

slide-17
SLIDE 17

Summary

  • A portrait painting method taking advantage of
  • Facial structure information
  • Techniques of artists
  • Future Work
  • Richer Representation
  • Better Color Processing
  • Part-Based Model
slide-18
SLIDE 18

http://bit.ly/portrait-painting

THANK YOU