MAT vs. CAT Maximal disc Any circle along with its interior points that is contained in the shape and touches the boundary at two or more points MAT Centers and radii of all maximal discs contained in the shape Problems with MAT (Pages 5-6) Indirect inversion Nonlocal transform Boundary feature may induce skeletal feature far away Boundary feature may be exaggerated or underplayed in skeleton Only info about distances of shape features, none about girths or sizes Hard to evaluate the significance of a feature Difficult to prune MAT to form "basic" skeleton MAT does not exist for discrete shape Requirements of skeleton for 2D planar shape: (i) connected 1D subset of shape (ii) maximally axial to shape (local axis of symmetry) (iii) Feature of skeleton should be nearby corresponding shape feature (iv) invertible (v) multiresolution (vi) applicable to discrete representations of shapes, coarser discretization should lead to coarse skeleton Maximal chord of tangency Chord of the bounding circle of a maximal disc CAT Midpoints and half-lengths of all maximal choords of tangency, and the centers and radii of a maximal disc with three maximal chords of tangency that form an acute angled triangle Differences between MAT and CAT MAT is connected CAT is disconnected protoskeleton Contiguous arcs are local axes of symmetry Isolated points are branch points -> Features of skeleton correspond to features of shape -> Size of shape feature corresponds to size of skeletal feature -> Protskeleton can be connected MAT is nonlocal Each pair/triple of boundary points can lead to infinite number of skeletal points (Figure 2e) CAT is local Each pair of boundary points leads to only one skeletal point and it is nearby and maximally axial Shape of CAT degrades gracefully with coarser discrete sampling MAT does not CAT stays along local symmetry axes MAT can go outside shape (Figure 3f) MAT is invertible CAT is strongly invertible Boundary can be recovered directly from CAT Each maximal chord is perpendicular to CAT axis Allows partial reconstruction CAT for discrete shapes Connect midpoints of delauney edges Prune based on d/|AB|, where d is distance from furthest point to edge AB Morphological congruency CAT provides attributed graph Grammar for shape description (J = T +2g -2) J = joints, S = sleeves, T = terminals