Transcript notes

Recognition of Shapes
by
Editing Shock Graphs
Thomas B. Sebastian
Philip N.Klein
Benjamin B. Kimia
Div. Of Engineering
Brown University
Dept. of Computer Science
Brown University
Div. Of Engineering
Brown University
Presented by Yuan Chen, Oct. 18, 2004
Outline
Shock Graph
 Key ideas
 Partitioning the Shape Space
 Deformation Path
 Edit Distance Algorithm
 Results

Shock Graph

Medial axis: locus of centers of maximal
circles that are bitangent to shape boundary.

Shock Graph
-- A variant of the medial axis
-- With direction of flow
-- Shock segment:
medial axis segment with monotonic flow;
a more refined partition of medial axis.
Shock Graph
Image courtesy of Thomas B. Sebastian et al, 2001.
Shock Transitions
• Boundaries
between shape cells ( collection of
shapes with same shock graph topology).
Image courtesy
of Thomas B.
Sebastian et al,
2003.
Key Ideas

All shapes with the same shock graph
topology are equivalent --- shape space
partitioned

All deformations with the same sequence
of shock graph transitions are equivalent.

The globally optimal transition sequence
can be found in polynomial time.
Partitioning the shape space

Shape space is
collection of all
shapes.
-- Shape is a point
-- Shape
deformation
sequence is a path.
-- Cost of optimal
deformation
sequence is
distance from A to
B.
Image courtesy of Thomas B. Sebastian et al, 2001.
Deformation Path


Sequence of shock graph transitions.
Shock graph transitions are formally
classified and the complete list is known:
Image
courtesy of
Thomas B.
Sebastian
et al, 2001.
Deformation Path

A shape deformation bundle is the set of
deformation paths passing through the
same set of shape cells. --- Discretization.
Image
courtesy of
Thomas B.
Sebastian et
al, 2001.
Deformation Path

Avoid complexity-increasing
deformation paths.
Image
courtesy of
Thomas B.
Sebastian et
al, 2001.
Edit Distance Algorithm

Apply the edit distance approach to
find the optimal path among all
possibilities. – polynomial time.
Image
courtesy of
Thomas B.
Sebastian et
al, 2001.
Four Groups of Edit Operations

Slice:

Contract:

Merge:

deletes a shock branch and merges the remaining two.
deletes a shock branch between degree-three nodes.
combines two branches at a degree-two node.
Deform:
relates two shapes in the same shape cell.
Images
courtesy of
Thomas B.
Sebastian et
al, 2001.
Assigning costs to edit operations

Derive the cost of the deform edit:
Sum over local shape differences –
differences between matching shock
segment attributes.

Other edit costs are considered to be
limiting cases of a deform cost.
Deform cost between shock segments




Length differences of the boundary segments
Curvature differences of boundary segments
Difference in width of shape
Difference in relative orientation of boundary
segments
Results

The performance of shock-graph matching in
the presence of commonly occurring visual
transformations:
-------

boundary perturbations
articulation and deformation of parts
shadow and highlights
viewpoint variation
scale
partial occlusion
Indexing results
Boundary perturbations
Shock computation is sensitive to boundary
perturbations. But the cost of a splice in this
case is very low.
Articulations and deformation of parts

Robust in the presence of
articulation and deformation of part
of the shape.
Shadow and Highlights
Shadow -- tend not to affect the
shock graph topology
 Highlight – tend to be small

Viewpoint variation

Smooth changes are handled by deform edit

Abrupt changes are handled by splices or
contract edit
Scale

Can’t be handled.
-- The shock-graph topology is not
changed by scale.
-- But the shock edge comparison is
not scale invariant!!
Partial Occlusion

Small partial occlusion can be
handled.
Indexing results

Handle small database very well:
9x11 and 8x12

To deal with large database:
-- improve the efficiency of matching
-- reduce the computational burden in
the indexing scheme
Thank You !