Patternbased Texture Metamorphosis Z Liu C Liu and
Pattern-based Texture Metamorphosis Z. Liu, C. Liu, and H. Shum Microsoft Research Asia Y. Yu UIUC
Image Morphing vs. Texture Morphing Image Morphing n n n Specify Features and Correspondence * Warp Generation Transition Control * Require a lot of human intervention
Image Morphing vs. Texture Morphing n Textures are usually homogenous with features everywhere. n n Hard to specify features Hard to build correspondence
Direct Blending Does Not Work source target Random Semi-structured Regular
Interesting Problems In Texture Morphing n What pair of textures? n n n What is the feature? n n User define pattern. How to extract so many patterns? n n Similar and repeatable patterns. Pattern distributions are alike. Semi-automatic approach. How to build correspondence? n Generate a smooth warp field.
Our Approach n n 1. Pattern Detection and Alignment 2. Establishing Correspondence ü ü Source texture n Target texture 3. Warping and Blending Morphing sequence
Pattern Representation & Distance Measurement n Pattern Representation n Shape Distance n Local Feature Distance
Pattern Detection & Alignment n Step 1: Initialization by Generalized Hough Transform (GHT). n Step 2: Alignment by top-down verification. n Step 3: Refinement by human intervention.
Pattern Detection & Alignment Step 1: Initialization Original texture Voting of a pixel Intensity image Local maximum User selected pattern
Pattern Detection & Alignment Step 2: Alignment n (a) Independently update each landmark n (b) Update shape n Iteratively do (a) and (b).
Alignment Process GHT initialization alignment
Pattern Detection & Alignment Step 3: Refinement n (a) False detection n (b) False alignment n (c) More than one types of pattern
Correspondence by Minimizing Morphing Path
Warping and Blending From S. Lee
More Results source target Pattern selected
Discussion n About Pattern Selection n About Correspondence and Transition Control n n Can be any shape User is responsible Problem of crowd patterns About Warp Generation n MFFD vs. “as rigid as possible”
Thank you !
- Slides: 17