Robust Near-Isometric Matching via Structured Learning of Graphical Models

M55

´ Julian J. McAuley, Tiberio S. Caetano, Alexander J. Smola We use structured learning in a graphical model to match shapes in images under near-isometric transformations.

our graphical model, and its embedding in an image

The result is an efficient and exact shape-matching algorithm, which includes first, second, and third-order features.

an example `shape' (left), and a match found in another image (right)