Back to Puzzle
Segmentation
Input: Photo of puzzle pieces
Output: List of (x,y) pairs for each puzzle piece
- Level set methods
- Yutian's ncut-derived method
- NJ's ideas from media class
Contour extraction / representation
Input: List of (x,y) pairs for each puzzle piece
Output: Some nicer representation? Perhaps a bunch of "piece objects"?
- Relative angles
- Discretised directions --> BLAST
- "Continuous" directions --> Correlation?
- Eamon's shape representation
- Alter slightly to accommodate comparisons
- Local polynomials, curvature
- Raman's n-gram approach
Comparison / local matching
- Techniques depend on representation
- Can include color, texture, etc.
Global solution / energy minimization
- Jon's approach for canonical puzzles
- Belief propagation / affinity propagation
- More ideas? Simple BFS?
Final reconstruction / assembly
- Stitch the pieces together so they look nice
