Pattern-based Texture Metamorphosis

1
Pattern-based Texture Metamorphosis

• Z. Liu, C. Liu, and H. Shum
• Microsoft Research Asia

Y. Yu
UIUC
2
Image Morphing vs. Texture Morphing
Image Morphing

• Specify Features and Correspondence
• Warp Generation
• Transition Control

Require a lot of human intervention
3
Image Morphing vs. Texture Morphing
Texture Morphing

• Textures are usually homogenous with features
  everywhere.
• Hard to specify features
• Hard to build correspondence

4
Direct Blending Does Not Work
source
target
Random
Semi-structured
Regular
5
Interesting Problems In Texture Morphing

• What pair of textures?
• Similar and repeatable patterns.
• Pattern distributions are alike.
• What is the feature?
• User define pattern.
• How to extract so many patterns?
• Semi-automatic approach.
• How to build correspondence?
• Generate a smooth warp field.

6
Our Approach

• 1. Pattern Detection and Alignment
• 2. Establishing Correspondence
• 3. Warping and Blending

Source texture
Target texture
Morphing sequence
7
Pattern Representation Distance Measurement

• Pattern Representation
• Shape Distance
• Local Feature Distance

8
Pattern Detection Alignment

• Step1 Initialization by Generalized Hough
  Transform (GHT).
• Step2 Alignment by top-down verification.
• Step3 Refinement by human intervention.

9
Step1 Initialization
Pattern Detection Alignment
Voting of a pixel
Original texture
User selected pattern
Intensity image
Local maximum
10
Step2 Alignment
Pattern Detection Alignment

• (a) Independently update each landmark
• (b) Update shape
• Iteratively do (a) and (b).

11
Alignment Process
GHT initialization
alignment
alignment
12
Pattern Detection Alignment
Step3 Refinement

• (a) False detection
• (b) False alignment
• (c) More than one types of pattern

13
Correspondence by Minimizing Morphing Path
14
Warping and Blending
From S.Lee
15
More Results
source
target
Pattern selected
16
Discussion

• About Pattern Selection
• Can be any shape
• User is responsible
• About Correspondence and Transition Control
• Problem of crowd patterns
• About Warp Generation
• MFFD vs. as rigid as possible

17
Thank you !